DigitalByte's picture
Add shoe first/last-frame example + workflow, workflows README and input frames; update model card, prompt guide and example workflow
0a7b464 verified
|
Raw
History Blame Contribute Delete
2.1 kB
# ComfyUI workflows
[Single-image PC workflow](ltx-2.5_distilled_xpldv_example_workflow.json) · [PC starting image](xpldv_computer.png) · [Recommended settings](../README.md#recommended-settings) · [Prompt guide](../prompt_guide.md)
## 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.
Try the running-shoe example from the demo reel. The workflow includes its exact tested prompt and applies both image guides in both sampling passes.
| First image: assembled | Last image: exploded |
| --- | --- |
| ![Assembled running shoe](xpldv_shoe_first.png) | ![Exploded running shoe](xpldv_shoe_last.png) |
1. Download the [shoe first/last-frame workflow](ltx-2.5_distilled_xpldv_first_last_shoe.json) and open it in ComfyUI.
2. Download the [first image](xpldv_shoe_first.png) and [last image](xpldv_shoe_last.png). Upload them in the matching image nodes, or copy both files into `ComfyUI/input`.
3. Select your installed model files and the released XPLDV LoRA. Keep the prefilled prompt and settings, then run the workflow.
This example uses XPLDV at **0.7**, seed **42**, and **145 frames at 24 FPS** (about six seconds), with a two-pass 960 × 544 → 1920 × 1088 render. Each endpoint guide is set to **1.0**. The exact prompt is filled in; the reusable BASIC template sits in a note above it.
[Watch the recorded shoe result](../examples/xpldv_shoe_first_last.mp4) · [First/last-frame prompting notes](../prompt_guide.md#first-and-last-frame-guidance)
The recorded result includes generated audio. Its end image guides the final stack, with some foam reshaping during separation. The single-image PC workflow above remains the simpler starting point at strength **0.5**.