Prophetero Scene Planner — Plan the scene. Keep the control.
You have the images. Now give the scene a plan.
Scene Planner brings image fitting, keyframe timing, scene prompts and LoRA choices into one large, local workspace for DaSiWa and Eros MiniMax H3 Simple/Pro workflows. It does not replace ComfyUI or generate a video itself: it prepares a new workflow copy for you to review and run.
What is inside?
- Up to six start, intermediate and end images, with ordering, timing and a large crop preview.
- Smaller trial sizes, portrait/landscape/square presets and custom dimensions, including 1088 × 608.
- Requested seconds beside actual H3 frame count and duration, so you can see what will really be exported.
- Editable timed scene events, copyable prompts and optional sound direction: ambience, effects, music or requested silence.
- Local LoRA search and supported loader editing: add, remove, switch and adjust strengths without repeatedly returning to the graph.
- A source-versus-planned review before export, plus scoped resets that preserve unrelated scene work.
- Saved projects, local draft recovery and optional local Ollama translation.
- Eight interface languages, English-first files and English/Hungarian beginner guides.
Getting started: extract the complete ZIP, stop your ComfyUI backend, run INSTALL.cmd, select the actual backend folder containing main.py, then fully restart SwarmUI/ComfyUI. Open the planner from your working H3 node or the canvas right-click menu. Read QUICKSTART_EN.md for a first short scene. Models, encoders, VAEs and required custom nodes remain separate downloads; see DEPENDENCIES.md.
This is a public beta, currently H3 only. LTX and arbitrary video architectures are not supported. Keyframes, approximate event times and sound direction guide the chosen model; they cannot guarantee exact motion or audio behavior. No actual audio-generation, lip-sync or automatic-transition engine is included. Start with a short trial on your own hardware.
Credits: Scene Planner's planning UI and workflow assembly are by Prophetero / ArtofTomworks. ComfyUI (https://github.com/Comfy-Org/ComfyUI), DaSiWa MiniMax H3 by Darksidewalker (https://civarchive.com/models/2877206), H3 Eros Max by tenstrip (https://civarchive.com/models/2851079), rgthree (https://github.com/rgthree/rgthree-comfy), KJNodes (https://github.com/kijai/ComfyUI-KJNodes) and Crystools (https://github.com/crystian/ComfyUI-Crystools) retain all credit for their own work. This utility is a companion to those tools and models; it does not train or redistribute them.
License: free use for personal and commercial creative work; free sharing/modification with the license retained. Resale or paid bundling of the program or modified versions is not allowed. Model and source-media licenses apply independently. Full terms are in LICENSE.txt.
If you try it, a short note about your source workflow and what made planning easier is very welcome. Reproducible bug reports help shape the next version.
Description
Changelog
0.2.0-beta.1 — 2026-11-11 — first public beta
Consolidates the private Scene Planner tests and fixes into one complete package.
- Large local planning page for DaSiWa/Eros H3 Simple and Pro source workflows; current-graph capture, manual import and four starting templates.
- Up to six start/guide/end images, ordering, timeline selection, large crop preview, padding/stretch options and automatic first-image orientation suggestions.
- Expanded trial/portrait/landscape/square/custom resolutions, including 1088 × 608.
- H3-aligned actual frames/FPS/duration display, explicit duration-control guidance and synchronization of recognized exclusively owned legacy controls.
- Timed scene events, editable/copyable prompt and optional textual sound direction.
- Local LoRA search/add/remove/toggle/strength controls for recognized loaders; compact enabled-only list, disabled-row recovery and scoped source reset.
- Source/planned comparison with labels on each card; source snapshot remains immutable and later ComfyUI edits require a fresh import.
- Local browser draft recovery, portable project JSON with images/source graph, and separate executable workflow export.
- Optional local Ollama translation with explicit review/apply; no model downloads in the installer.
- Prepared-image recognition in ComfyUI, missing-source recovery and conservative rejection of ambiguous loader/graph layouts.
- Eight UI languages, English-first release files, English/Hungarian beginner instructions, package hashes and Windows installer backups.
Scope: H3 only. LTX is postponed. No generation engine, automatic queueing, actual audio synthesis/repair, automatic transitions or lip-sync engine is included.
Validation distinguishes isolated code/browser/installer checks from actual model generation. No general video-quality or hardware-compatibility guarantee is claimed.