Add first/last-frame guidance, unwrap templates, refresh prompt guide

Uploaded via huggingface_hub API. Workflow and training_config.yaml already match the remote.

README.md

@@ -57,6 +57,12 @@ Turn a source image into an animated exploded view. XPLDV separates an object's
 
 XPLDV works best with `ltx-2.5-22b-distilled-transformer-bf16.safetensors`. It can also be used with `ltx-2.5-22b-dev-transformer-bf16.safetensors`, which it was trained on. The included workflow uses the distilled model. Base models: [Lightricks/LTX-2.5](https://huggingface.co/Lightricks/LTX-2.5).
 
+## First and last frame guidance
+
+First/last-frame generation also works well with XPLDV. Supply the assembled subject as the first image and a matching exploded arrangement as the last image. **In this setup, XPLDV provides motion guidance for the separation between the two images.** The endpoint images establish the starting appearance and intended final arrangement; the prompt describes which parts detach, their paths, and the ongoing subject and camera movement.
+
+This approach worked well for the burger and running-shoe reveals in the demo reel. Use a workflow that accepts both image guides. The included example workflow uses one starting image; see the [first/last-frame setup and prompting notes](prompt_guide.md#first-and-last-frame-guidance) for the tested two-guide setup.
+
 ## Featured videos
 
 Six 10-second examples using XPLDV at strength **0.5**. Each video includes its exact prompt.
@@ -107,31 +113,20 @@ Change one setting at a time while keeping the image, prompt, and seed fixed.
 
 Start with **"separates into an ordered exploded view."** This phrase appears in all 204 separation training captions. Name the parts that move, the components revealed underneath, and the final arrangement. Write one continuous paragraph.
 
+Use the [slot reference](prompt_guide.md#filling-the-slots) to fill the placeholders in either template.
+
 “Separates into individual components” is another phrasing we tested. With either opening, describe the movement and the final arrangement.
 
 ### BASIC template — a short, readable reveal
 
 ```text
-[SUBJECT] separates into an ordered exploded view. [NAMED OUTER PARTS] move
-outward along controlled paths, revealing [NAMED INTERNAL COMPONENTS]
-underneath. The parts settle into an organized [LAYERED / RADIAL]
-arrangement around [EXPOSED CORE] while retaining their original shapes.
-[CAMERA BEHAVIOR]. [LIGHTING CONSISTENT WITH THE IMAGE] remains consistent
-throughout the continuous shot.
+[SUBJECT] separates into an ordered exploded view. [NAMED OUTER PARTS] move outward along controlled paths, revealing [NAMED INTERNAL COMPONENTS] underneath. The parts settle into an organized [LAYERED / RADIAL] arrangement around [EXPOSED CORE] while retaining their original shapes. [CAMERA BEHAVIOR]. [LIGHTING CONSISTENT WITH THE IMAGE] remains consistent throughout the continuous shot.
 ```
 
 #### Example: Desktop PC
 
 ```text
-The desktop computer tower separates into an ordered exploded view. The
-glass side panel, front panel, top cover and lower power-supply cover move
-outward along controlled paths, revealing the motherboard, graphics card,
-liquid-cooling assembly and power supply. The components separate into an
-organized layered arrangement around the case while retaining their original
-shapes. The case base stays on the desk. The camera slowly orbits to the
-right around the computer from the starting three-quarter view. The
-blue-white computer lighting and surrounding room remain consistent
-throughout the continuous shot.
+The desktop computer tower separates into an ordered exploded view. The glass side panel, front panel, top cover and lower power-supply cover move outward along controlled paths, revealing the motherboard, graphics card, liquid-cooling assembly and power supply. The components separate into an organized layered arrangement around the case while retaining their original shapes. The case base stays on the desk. The camera slowly orbits to the right around the computer from the starting three-quarter view. The blue-white computer lighting and surrounding room remain consistent throughout the continuous shot.
 ```
 
 [Watch the example](examples/xpldv_computer.mp4)
@@ -141,54 +136,18 @@ throughout the continuous shot.
 A COMPLEX prompt guides a more detailed, cinematic reveal through successive layers, describing deeper internal components, their appearance, and their positions, including details hidden or difficult to see in the starting image. These descriptions give LTX 2.5 the information needed to construct the intended reveal beyond what the image alone shows.
 
 ```text
-[SUBJECT] separates into an ordered exploded view while [CONTINUING ACTION
-OR ANCHORED SUPPORT]. [OUTER PART A] detaches and translates [DIRECTION],
-keeping [ORIENTATION]. [OUTER PART B] moves [DISTINCT PATH], and [OTHER
-OUTER PARTS] move [PATHS]. These exterior parts move first and travel
-furthest, leaving clear gaps. They reveal [SPECIFIC COMPONENT AND
-RECOGNIZABLE FEATURES] at [POSITION RELATIVE TO THE OBJECT], with [OTHER
-COMPONENTS AND THEIR LOCATIONS]. [CORE COMPONENTS] remain mounted together.
-Each detached section retains its shape as the arrangement settles and holds
-steady. [CAMERA BEHAVIOR]. [CONSISTENT LIGHTING] continues through one
-continuous shot.
+[SUBJECT] separates into an ordered exploded view while [CONTINUING ACTION OR ANCHORED SUPPORT]. [OUTER PART A] detaches and translates [DIRECTION], keeping [ORIENTATION]. [OUTER PART B] moves [DISTINCT PATH], and [OTHER OUTER PARTS] move [PATHS]. These exterior parts move first and travel furthest, leaving clear gaps. They reveal [SPECIFIC COMPONENT AND RECOGNIZABLE FEATURES] at [POSITION RELATIVE TO THE OBJECT], with [OTHER COMPONENTS AND THEIR LOCATIONS]. [CORE COMPONENTS] remain mounted together. Each detached section retains its shape as the arrangement settles and holds steady. [CAMERA BEHAVIOR]. [CONSISTENT LIGHTING] continues through one continuous shot.
 ```
 
 #### Example: Watch
 
 ```text
-A luxury automatic mechanical wristwatch separates into an ordered exploded
-view while resting stationary on its dark velvet display cushion. The front
-sapphire crystal and surrounding polished bezel rim detach as a single unit
-and translate forward and slightly upward, keeping its circular orientation
-facing the camera. The hour, minute, and slender seconds hands lift cleanly
-above the dial plane, each translating upward along its own short vertical
-path with slight lateral spacing between them. The main dial plate carrying
-its applied hour markers, subdial components, and open-heart cutout details
-moves forward a shorter distance than the crystal, creating a clear gap
-between it and the hands above. These exterior parts move first and travel
-furthest, leaving generous gaps between each successive layer. They reveal
-the mechanical movement at the center of the stack—its polished bridges,
-interlocking gears, escapement fork, oscillating balance wheel, ruby jewels,
-and complication modules all visible in their assembled configuration—with
-the rear exhibition caseback and its oscillating automatic rotor translating
-backward and slightly downward as the final layer. The movement's bridges,
-gears, and escapement remain mounted together as a single mechanical unit.
-Each detached section retains its original shape as the five layers settle
-into an organized axial arrangement and hold steady. A slow, steady camera
-pull-back provides room for the full expanded stack. The soft upper-left key
-light, cool right-side fill, and subtle rim highlight consistent with the
-source image continue through one continuous shot.
+A luxury automatic mechanical wristwatch separates into an ordered exploded view while resting stationary on its dark velvet display cushion. The front sapphire crystal and surrounding polished bezel rim detach as a single unit and translate forward and slightly upward, keeping its circular orientation facing the camera. The hour, minute, and slender seconds hands lift cleanly above the dial plane, each translating upward along its own short vertical path with slight lateral spacing between them. The main dial plate carrying its applied hour markers, subdial components, and open-heart cutout details moves forward a shorter distance than the crystal, creating a clear gap between it and the hands above. These exterior parts move first and travel furthest, leaving generous gaps between each successive layer. They reveal the mechanical movement at the center of the stack—its polished bridges, interlocking gears, escapement fork, oscillating balance wheel, ruby jewels, and complication modules all visible in their assembled configuration—with the rear exhibition caseback and its oscillating automatic rotor translating backward and slightly downward as the final layer. The movement's bridges, gears, and escapement remain mounted together as a single mechanical unit. Each detached section retains its original shape as the five layers settle into an organized axial arrangement and hold steady. A slow, steady camera pull-back provides room for the full expanded stack. The soft upper-left key light, cool right-side fill, and subtle rim highlight consistent with the source image continue through one continuous shot.
 ```
 
 [Watch the example](examples/xpldv_watch.mp4)
 
-Keep related parts within a named internal mechanism intact when useful, such as the gears inside a watch movement. This does not keep the exterior attached. Describe which internal layers also move; omit the sentence about components remaining mounted together when it does not fit the reveal.
-
-Replace the brackets with your subject and its parts; omit clauses that do not apply to the intended shot. Name the final arrangement: a vertical stack, a horizontal row, or a radial arrangement around the center. Use “layered” for an actual stack. Describe timing consistently: parts move at the same time or in a clear sequence. For a moving subject, keep the detached parts traveling with it. Give the camera enough room for the expanded arrangement.
-
-Audio is optional. Add sound effects, ambience, music, or dialogue in a final sentence and keep the audio pipeline connected. Audio wording can influence the visual action.
-
-See the [prompt guide](prompt_guide.md) for a short walkthrough and troubleshooting tips.
+See the prompt guide for [motion, preservation, and optional audio](prompt_guide.md#keep-the-shot-consistent), plus [troubleshooting](prompt_guide.md#when-the-result-needs-work).
 
 ## Limitations and tips
 

prompt_guide.md

@@ -1,43 +1,73 @@
 # Prompting XPLDV
 
-Start with **“separates into an ordered exploded view.”** Name what moves, what becomes visible, and where the parts finish. Use direct words such as **moves up**, **moves outward**, and **remains suspended**.
+Start with **“separates into an ordered exploded view.”** Choose a [BASIC or COMPLEX template](README.md#prompting), then use the slot reference below to fill it.
 
-“Separates into individual components” is another phrasing we tested. With either opening, describe the movement and the final arrangement.
+## Build the reveal
 
-Use the [BASIC and COMPLEX templates](README.md#prompting) to start. A BASIC prompt names the main parts and the intended reveal.
+### Filling the slots
 
-A COMPLEX prompt guides a more detailed, cinematic reveal through successive layers, describing deeper internal components, their appearance, and their positions, including details hidden or difficult to see in the starting image. These descriptions give LTX 2.5 the information needed to construct the intended reveal beyond what the image alone shows.
+Grouped names share an explanation. Fill only the slots in your chosen template, giving each part its own details, then write the finished prompt as one paragraph.
 
-The [featured videos](README.md#featured-videos) contain complete prompts, and the [recommended settings](README.md#recommended-settings) are in the README.
+| Slot | What goes in it |
+| --- | --- |
+| `[SUBJECT]` | The thing being taken apart: “the black SUV.” |
+| `[NAMED OUTER PARTS]`<br>`[OUTER PART A]`<br>`[OUTER PART B]`<br>`[OTHER OUTER PARTS]` | Outside pieces to detach: “hood, front bumper, fenders.” Use the individual part slots when filling COMPLEX. |
+| `[NAMED INTERNAL COMPONENTS]` | Inside parts to reveal: “engine, radiator, and front seats.” |
+| `[LAYERED / RADIAL]` | Stacked layers or parts around a center. Example: “radial.” |
+| `[EXPOSED CORE]` | The center the parts surround: “the moving chassis.” |
+| `[CAMERA BEHAVIOR]` | One camera action with room for all the parts: “The camera tracks alongside at matching speed.” |
+| `[LIGHTING CONSISTENT WITH THE IMAGE]`<br>`[CONSISTENT LIGHTING]` | The light already in the image: “the warm street lighting.” |
+| `[CONTINUING ACTION OR ANCHORED SUPPORT]` | What the subject keeps doing or rests on: “driving across the bridge.” |
+| `[DIRECTION]`<br>`[DISTINCT PATH]`<br>`[PATHS]` | Each part's movement path: “hood straight upward; bumper forward.” Give each part its own path. |
+| `[ORIENTATION]` | How it stays angled: “flat and horizontal.” |
+| `[SPECIFIC COMPONENT AND RECOGNIZABLE FEATURES]` | An internal part and its appearance: “a V8 engine with exposed intake pipes.” |
+| `[POSITION RELATIVE TO THE OBJECT]` | Where that part belongs: “ahead of the cabin.” |
+| `[OTHER COMPONENTS AND THEIR LOCATIONS]` | Other revealed assemblies and their positions: “brake discs inside the wheels.” |
+| `[CORE COMPONENTS]` | Parts to keep assembled: “the engine and transmission.” Omit this clause if those parts should separate too. |
 
-## Build the reveal
+## Keep the shot consistent
 
-1. **Choose the subject.** Describe the main object in the image. If several objects are visible, identify yours by its position or appearance.
-2. **Name what separates.** Each exterior part selected for separation must come completely free, move away, and remain suspended with a visible gap from the subject. Name parts individually when they should detach independently.
-3. **Name what becomes visible.** Be specific about the important interior parts and where they belong: “the V6 engine block ahead of the cabin” or “the battery beneath the palm rest.” Add recognizable features only when they help describe the reveal.
-4. **Describe the finish.** State that the detached parts remain suspended with visible gaps and keep their original shapes. Describe which internal layers also move and which remain in place.
+State ongoing motion separately from separation: the vehicle keeps driving, its wheels keep rotating, and the detached parts keep traveling with it after the gaps stop increasing. Keep stationary subjects on their support.
 
-Match the final arrangement to the movement:
+Match the paths to the final arrangement: vertically apart for a stack, sideways for a row, or outward around the core for a radial spread. Describe one consistent order, whether parts separate together or in stages. Detached parts should keep their shapes and remain suspended with visible gaps.
 
-| Movement | Example wording |
-| --- | --- |
-| Vertical stack | “The screen, frame, and circuit board move vertically apart and remain suspended in aligned layers with visible gaps.” |
-| Horizontal row | “The components move horizontally apart and remain suspended in an ordered row.” |
-| Radial | “The panels move outward from the frame in a radial arrangement and remain suspended with visible gaps.” |
+**Optional additions**
 
-Use “layered” for a stack. Different parts can move in different directions: a hood moves upward while doors move sideways. Add orientation details when they matter, such as keeping a glass cover flat.
+- **Preservation:** Name existing people, objects, or surroundings that should stay consistent. For example: “The driver remains seated and the bridge stays intact.” These are preservation requests, not guaranteed locks.
+- **Audio:** With audio generation enabled, add one final sentence for sound effects, ambience, music, or dialogue. Put spoken dialogue in quotation marks. Audio wording can influence the visible action, so review picture and sound together. See [LTX's prompting guide](https://ltx.io/blog/ltx-2-5-prompt-guide) for examples.
 
-Describe one consistent timing pattern. If parts start together, say they move apart simultaneously. If they separate in stages, describe that order. Do not combine “first, then” with a requirement that every part start together.
+## First and last frame guidance
 
-Keep related parts within a named internal mechanism intact when useful, such as the gears inside a watch movement. This does not keep the exterior attached. Describe which internal layers also move; omit the sentence about components remaining mounted together when it does not fit the reveal.
+XPLDV also works well with first/last-frame generation. **For this setup, use XPLDV as motion guidance between the two endpoint images.** The first image establishes the assembled subject; the last image defines the intended exploded arrangement. XPLDV encourages ordered component separation during the movement between them.
 
-## Keep the shot consistent
+1. **Prepare a compatible image pair.** Keep the same subject, identifiable parts, materials, and lighting. Give the last image clear gaps between components and enough room for the full arrangement. Any change in viewpoint should fit the camera movement you request.
+2. **Use a workflow that accepts both guides.** The tested ComfyUI setup uses `LTXVAddGuide` at frame `0` and frame `-1`, with both guides applied in each sampling pass. The included release workflow is a single-image example; it needs a first/last-frame variant for this setup.
+3. **Describe the movement between the images.** Use the existing BASIC or COMPLEX structure. Name the parts, their detachment paths, what becomes visible, and what continues moving. Write a consistent route to the supplied final arrangement.
+4. **Review the whole transition.** Check component identity, detachment, and continuous movement between the endpoints. A matching last frame alone does not establish a clean exploded-view animation.
 
-Match the source image's motion and setting. A stationary object stays on its support; a moving vehicle keeps traveling, with its detached parts traveling alongside it.
+An end image can strongly determine the final geometry. To judge XPLDV's contribution, compare the movement between endpoints with the LoRA enabled and disabled while keeping both images, prompt, seed, and sampling settings fixed.
 
-Choose one camera movement. Use a fixed view when there is already room for the reveal, a gentle pullback when the parts need more space, or a tracking shot for a moving subject. Request an orbit only when you want one. Keep the source lighting consistent.
+<details>
+<summary>Tested first/last-frame settings</summary>
+
+The burger and running-shoe tests used the following settings. These record the successful guided tests; **0.5 remains the general starting LoRA strength for a new subject**.
+
+| Setting | Tested value |
+| --- | --- |
+| Model | LTX-2.5 distilled BF16 |
+| XPLDV | Released step 3000 weights, strength 0.7 in both passes |
+| First and last image guides | Strength 1.0 at frame 0 and frame -1 in both passes |
+| Duration | 145 frames at 24 FPS, approximately 6 seconds |
+| First pass | 960 x 544; Euler ancestral; 8 steps |
+| First-pass manual sigmas | `1.0, 0.99375, 0.9875, 0.98125, 0.975, 0.909375, 0.725, 0.421875, 0.0` |
+| Second pass | 1920 x 1088; Euler; 3 steps |
+| Second-pass manual sigmas | `0.85, 0.7250, 0.4219, 0.0` |
+| CFG and seed | CFG 1 and seed 42 in both passes |
+| Text conditioning | Direct prompt, enhancement off, empty negative prompt |
+
+LoRA strength and image-guide strength are separate controls. Keep both passes consistent when comparing LoRA strengths. Review each new image pair; the tested strength is not a guarantee of better quality for another subject.
 
-Audio is optional. Add a short final sentence for sound effects, ambience, music, or dialogue; put spoken words in quotation marks. For example: “Soft mechanical clicks accompany the separation over quiet room ambience.” Audio wording can influence the action, so check the picture and sound together. See [LTX's prompting guide](https://ltx.io/blog/ltx-2-5-prompt-guide) for more audio examples.
+</details>
 
 ## When the result needs work
 
@@ -52,10 +82,9 @@ Audio is optional. Add a short final sentence for sound effects, ambience, music
 
 Change one thing at a time. Keep the image, seed, and settings fixed when testing a prompt revision. Watch the whole clip: a good final frame can hide problems during separation. Detailed-looking internals may still be inaccurate.
 
-<details>
-<summary>Qwen prompt builder — copy with your image</summary>
+## Qwen prompt builder — copy with your image
 
-With an Ollama node, supply the builder below as the instruction template and enter your shot request in **Guidance**. Name the action, BASIC or COMPLEX, the separation direction, and the parts that should separate. For motion throughout the clip, say so explicitly in Guidance and check that Qwen includes it in the generated prompt.
+With an Ollama node, supply the image and put the builder below in the connected **Instruction Template** input. **Guidance** is optional: leave it blank for BASIC or enter **Complex** for a detailed reveal. Use Guidance to specify a particular subject, action, arrangement, or parts. Start a fresh conversation for each independent image. For motion throughout the clip, say so explicitly in Guidance and check that Qwen includes it in the generated prompt.
 
 Example Guidance:
 
@@ -66,90 +95,32 @@ Review both motion and separation in the finished clip. If your assistant has no
 ```text
 Write one LTX 2.5 image-to-video prompt for the XPLDV exploded-view LoRA.
 
-Inspect the supplied image. Identify the target subject, its current motion or
-support, the exterior components that should detach, the desired internal
-reveal, the structure that stays assembled, appropriate movement paths, camera
-framing and lighting continuity. Follow the shot request supplied in Guidance,
-including its action, mode, and named parts.
-
-Before writing the final prompt, privately plan five major assemblies or
-reveal stages based on the subject's construction. Identify the outer parts
-that detach, the deeper components revealed beneath them, and which internal
-mechanisms remain intact. Plan how each layer moves or becomes exposed, using
-the natural exterior-to-interior, front-to-back, or top-to-bottom order of the
-subject. Focus on large, visually meaningful assemblies and group related
-internal components when they belong together. This hierarchy guides the
-reveal; it does not require five detached pieces or a fixed animation
-sequence. Do not output the plan.
-
-Use BASIC for quick readable separation with a short list of named internal
-components. A COMPLEX prompt guides a more detailed, cinematic reveal through
-successive layers, describing deeper internal components, their appearance,
-and their positions, including details hidden or difficult to see in the
-starting image. These descriptions give LTX 2.5 the information needed to
-construct the intended reveal beyond what the image alone shows. Honor the
-requested mode; if none is specified, start with BASIC.
+Inspect the whole image and decide what physical construction the scene represents before naming parts. Choose the subject from the overall composition, not a prominent label or logo on one component. When a scene is built inside a larger machine or structure, use that enclosing construction as the subject and include its major visible levels; do not reduce the effect to a smaller object nested inside it. For a self-contained product on a table or in a presentation box, use the product itself; unrelated packaging and display props are not part of its construction. An explicit target or exclusion in Guidance overrides this default. State the chosen scope clearly in the opening sentence and keep every layer and internal reveal within that scope, preserving the relative scale and positions of its parts.
+
+Identify the subject's current motion or support, the exterior components that should detach, the desired internal reveal, the structure that stays assembled, appropriate movement paths, camera framing and lighting continuity. Use only the current image and shot request; do not carry subject restrictions or scene actions over from earlier requests. Follow Guidance for the action, mode, and named parts. Preserve the people and loose props while following any requested action. Where people or furniture need support, identify the load-bearing surface that stays in place and separate removable sections around it. Do not detach that same support elsewhere in the prompt.
+
+Before writing the final prompt, privately plan five major assemblies or reveal stages based on the subject's construction. Identify the outer parts that detach, the deeper components revealed beneath them, and which internal mechanisms remain intact. Plan how each layer moves or becomes exposed, using the natural exterior-to-interior, front-to-back, or top-to-bottom order of the subject. Focus on large, visually meaningful assemblies and group related internal components when they belong together. This hierarchy guides the reveal; it does not require five detached pieces or a fixed animation sequence. Do not output the plan.
+
+Use BASIC for quick readable separation with a short list of named internal components. A COMPLEX prompt guides a more detailed, cinematic reveal through successive layers, describing deeper internal components, their appearance, and their positions, including details hidden or difficult to see in the starting image. These descriptions give LTX 2.5 the information needed to construct the intended reveal beyond what the image alone shows. Honor the requested mode; if none is specified, start with BASIC.
 
 BASIC TEMPLATE:
-[SUBJECT] separates into an ordered exploded view. [NAMED OUTER PARTS] move
-outward along controlled paths, revealing [NAMED INTERNAL COMPONENTS]
-underneath. The parts settle into an organized [LAYERED / RADIAL] arrangement
-around [EXPOSED CORE] while retaining their original shapes. [CAMERA
-BEHAVIOR]. [LIGHTING CONSISTENT WITH THE IMAGE] remains consistent throughout
-the continuous shot.
+[SUBJECT] separates into an ordered exploded view. [NAMED OUTER PARTS] move outward along controlled paths, revealing [NAMED INTERNAL COMPONENTS] underneath. The parts settle into an organized [LAYERED / RADIAL] arrangement around [EXPOSED CORE] while retaining their original shapes. [CAMERA BEHAVIOR]. [LIGHTING CONSISTENT WITH THE IMAGE] remains consistent throughout the continuous shot.
 
 COMPLEX TEMPLATE:
-[SUBJECT] separates into an ordered exploded view while [CONTINUING ACTION OR
-ANCHORED SUPPORT]. [OUTER PART A] detaches and translates [DIRECTION], keeping
-[ORIENTATION]. [OUTER PART B] moves [DISTINCT PATH], and [OTHER OUTER PARTS]
-move [PATHS]. These exterior parts move first and travel furthest, leaving
-clear gaps. They reveal [SPECIFIC COMPONENT AND RECOGNIZABLE FEATURES] at
-[POSITION RELATIVE TO THE OBJECT], with [OTHER COMPONENTS AND THEIR
-LOCATIONS]. [CORE COMPONENTS] remain mounted together. Each detached section
-retains its shape as the arrangement settles and holds steady. [CAMERA
-BEHAVIOR]. [CONSISTENT LIGHTING] continues through one continuous shot.
-
-Adapt the chosen template to the actual image. Replace every placeholder with
-concrete wording. Keep BASIC concise. In COMPLEX, describe the successive
-layers, their appearance and positions, and the movement paths that expose
-them. Keep related parts within a named internal mechanism intact only when
-useful; this does not keep exterior panels attached. Describe which internal
-layers also move, and omit the sentence about components remaining mounted
-together when it does not fit the reveal. Add small parts only when they
-matter to the requested reveal. Do not require the same component to stay
-attached and detach independently. When full separation is requested, describe
-the part coming free of its attachments and remaining suspended with a visible
-gap; opening on a hinge is not full separation.
-
-Preserve continuing motion when requested or clearly part of the intended
-shot, with the separated components traveling with the subject. Keep that
-action throughout the shot, including the final arrangement. Holding the
-arrangement steady means keeping the gaps steady relative to the moving
-subject. Keep stationary subjects on their support. Choose one camera behavior
-with enough room for the expanded arrangement and retain the source lighting.
-
-Describe clear spatial and assembly relationships. Use the selected template
-to describe the final arrangement. Keep "separates into an ordered exploded
-view" as the default opening; use "separates into individual components" when
-requested or testing that alternative. Match the arrangement to the paths: a
-vertical stack, horizontal row, or radial arrangement around the core. Adapt
-the exterior-first timing when a different sequence is requested. Keep every
-major part requested in Guidance; do not substitute smaller decorative details
-for it.
-
-When audio is requested, append a concise final sentence describing the
-desired sound effects, ambience, music or dialogue. Quote spoken dialogue.
-Match the audio to the intended visual action, recognizing that audio wording
-may influence that action. When audio is not requested, omit this optional
-sentence.
-
-Do not add destruction, reassembly, an orbit, a negative-prompt list or
-unrequested scene changes by default. Do not invent a required trigger token.
-Do not promise perfect part counts, background isolation or accurate hidden
-construction.
-
-Return only the completed positive prompt as one continuous paragraph, without
-headings, alternatives, placeholders or explanations.
+[SUBJECT] separates into an ordered exploded view while [CONTINUING ACTION OR ANCHORED SUPPORT]. [OUTER PART A] detaches and translates [DIRECTION], keeping [ORIENTATION]. [OUTER PART B] moves [DISTINCT PATH], and [OTHER OUTER PARTS] move [PATHS]. These exterior parts move first and travel furthest, leaving clear gaps. They reveal [SPECIFIC COMPONENT AND RECOGNIZABLE FEATURES] at [POSITION RELATIVE TO THE OBJECT], with [OTHER COMPONENTS AND THEIR LOCATIONS]. [CORE COMPONENTS] remain mounted together. Each detached section retains its shape as the arrangement settles and holds steady. [CAMERA BEHAVIOR]. [CONSISTENT LIGHTING] continues through one continuous shot.
+
+Adapt the chosen template to the actual image. Replace every placeholder with concrete wording. Keep BASIC concise. In COMPLEX, describe the successive layers, their appearance and positions, and the movement paths that expose them. Keep related parts within a named internal mechanism intact only when useful; this does not keep exterior panels attached. Describe which internal layers also move, and omit the sentence about components remaining mounted together when it does not fit the reveal. Add small parts only when they matter to the requested reveal. Do not require the same component to stay attached and detach independently. When full separation is requested, describe the part coming free of its attachments and remaining suspended with a visible gap; opening on a hinge is not full separation.
+
+Preserve continuing motion when requested or clearly part of the intended shot, with the separated components traveling with the subject. Keep that action throughout the shot, including the final arrangement. Holding the arrangement steady means keeping the gaps steady relative to the moving subject. Keep stationary subjects on their support. Choose one camera behavior with enough room for the expanded arrangement and retain the source lighting.
+
+Describe clear spatial and assembly relationships. Use the selected template to describe the final arrangement. Keep "separates into an ordered exploded view" as the default opening; use "separates into individual components" when requested or testing that alternative. Match the arrangement to the paths: a vertical stack, horizontal row, or radial arrangement around the core. Adapt the exterior-first timing when a different sequence is requested. Keep every major part requested in Guidance; do not substitute smaller decorative details for it.
+
+When audio is requested, append a concise final sentence describing the desired sound effects, ambience, music or dialogue. Quote spoken dialogue. Match the audio to the intended visual action, recognizing that audio wording may influence that action. When audio is not requested, omit this optional sentence.
+
+Do not add destruction, reassembly, an orbit, a negative-prompt list or unrequested scene changes by default. Do not invent a required trigger token. Do not promise perfect part counts, background isolation or accurate hidden construction.
+
+Before returning, check that the completed prompt keeps the chosen subject and scope throughout, that no supporting part is described as both fixed and moving, and that separated parts remain suspended. Correct any conflicts silently.
+
+Return only the completed positive prompt as one continuous paragraph, without headings, alternatives, placeholders or explanations.
 ```
 
-</details>