# ArtofTomworks Scene DNA 0.7.0 — Plan the scene. Follow every shot.
A good sequence needs more than a good individual prompt. Scene DNA keeps your location, subject, appearance, lighting and visual direction together while you plan the moments around them.
This standalone, local creative companion brings together a shared scene recipe, 3–6 editable shots, reference images, composition planning, separate image/video prompts and a director’s overview. Lock the details that should stay consistent, vary the ones that should change, and keep the project for your next session.
Optional local AI can suggest an editable scene from your reference image, offer three alternatives and translate prompts beside the copy buttons. Choose your own installed Ollama model; AI and model downloads are not bundled. Attaching an image alone does not send an AI request.
New in 0.7: visible shot execution. With your installed, supported H3 Simple/Pro ComfyUI workflow, prepare individual shots or run selected shots one after another. The live monitor shows the actual shot, start image, dimensions, frame-derived duration, current node/steps and elapsed time. Existing ComfyUI/KJ previews appear here in a large view; completed videos, saved locations and downloads stay close to the work. Pause, explicit retry and missing-shot resume are included.
You can still export a separate ComfyUI workflow JSON for each shot instead. Preparing a project never starts generation by itself. Source model/LoRA/sampler/seed settings are retained in execution copies; imported source settings are snapshots. Important: unchanged completed shots are reused from their original folders. They are not regenerated or copied into the new run folder. Fewer files in that newest folder do not necessarily mean a missing shot. The UI flags reused results/multiple folders, and every result card shows its actual saved path and individual download.
Choose your package:
- Standard — recommended for creators already comfortable with local creative tools and ComfyUI workflows. Includes the complete application, examples and reference guides.
- Beginners — recommended if this is your first storyboard or Scene DNA project, you are new to ComfyUI workflow handoff, or you simply prefer a guided first session. Includes the same features, plus START_HERE and a step-by-step first-project walkthrough. It is not a reduced edition and does not install the required generation models for you.
Extract the full ZIP. Offline planning needs a modern desktop browser (START_OFFLINE.cmd). Local AI/server and the shot runner require separately installed Node.js 22+ (START.cmd). Optional image analysis also needs running Ollama plus an installed vision-capable model. The optional included ComfyUI bridge adds source capture/scoped stop; its installer asks for the actual backend folder containing main.py. Restart the backend after installing that bridge.
Not limited to H3 for prompt planning: use the scene recipes and copied prompts with your preferred generator. The built-in execution runner currently supports recognized H3 Simple/Pro branches only. LTX/arbitrary workflows are not supported by this runner. Any H3 base-model label on this listing refers to that execution support, not an exclusive requirement for the planner.
Eight interface languages. Windows/Brave tested; other platforms are not certified. Some newest monitor/diagnostic labels fall back to English outside HU/EN. No models, personal media, private projects or credentials are included. The cover is promotional artwork, not a screenshot.
This is a public beta. Composition previews and natural-language direction are planning aids, not guaranteed camera/motion/identity control. There is no automatic shot joining, transition editor or lip-sync. Heavy local AI and video generation should not compete for a memory-limited GPU. Save project files for backup, and restart Scene DNA’s own server after an update; browser refresh alone cannot update server code.
Planning, companion bridge and monitoring: ArtofTomworks. ComfyUI, Ollama, Node.js, generation-model authors and third-party workflow/node creators retain their own credits and terms; see THIRD_PARTY.md. Free personal/commercial creative use; free sharing/modification with license retained; no resale or paid bundling of this software or modified versions. Full terms: LICENSE.txt.
GitHub, source, guides and downloads: https://github.com/artoftomworks-debug/artoftomworks-scene-dna
Try one modest-resolution short shot first. If you report an issue, include the version, workflow type and exact error without private material. Feedback about confusing buttons or missing explanations is especially welcome.
pls read this guide: https://civarchive.com/articles/38739/scene-dna
Description
Changelog
0.7.0 — 12 November 2026 — Public Beta
First public Scene DNA package, consolidating the local 0.1–0.7 development previews.
New since the 0.6 planning preview
Optional local H3 Simple/Pro shot runner: captured workflow source or API JSON, selected output branch, one/several/all-shot preparation, separate JSON/ZIP export and explicit sequential execution.
Source model, LoRA, sampler and seed preservation; editable dimensions, optional original start image, short/full duration and displayed actual H3 frame-derived timing.
Start-image orientation detection before preparation; importing a source no longer silently overwrites the selected dimensions. Manual override remains available.
Explicit original-versus-translated prompt choice and visible preparation errors naming the affected shot.
Live monitor in Scene DNA: actual running shot/start image, node/steps/elapsed time, same-client ComfyUI and KJ still/WebP/MP4 previews, large view, floating return and scoped run controls.
Read-only exact submitted-node/connection overview and workflow export. The native ComfyUI canvas is not replaced automatically.
Completed full video playback in the monitor, saved output location and download. Reused shots are explicitly labelled, including older run records; a summary flags results spanning multiple output folders. Results are already saved automatically by ComfyUI.
Persisted run status, explicit failure retry and missing-shot resume; unchanged completed shots reused when expanding a selection. No automatic submission on page/project load.
Clear bounded execution-error extraction, dropping large tensor dumps. Out-of-memory and lost-connection states remain visible instead of being treated as successful completion.
Image-state startup race fixed: full browser image restoration precedes auto-inspection/autosave. Project-file backup remains recommended.
Packaged startup/runtime discovery, optional bridge installer with backend-folder selection, complete English-first Standard and Beginners ZIPs, bilingual guides and explicit beta limits.
Included planning features from 0.1–0.6
Shared, lockable scene recipe; 3–6 editable shots, camera/framing/movement, per-shot changes, image and video prompts, sound direction, storyboard/PDF and prompt-sheet export.
Large composition planner, draggable crops, wheel zoom, focus point and optional start/end framing.
Expandable lighting/style/palette/sound recipes, custom libraries and per-prompt local translation.
Optional image-aware local AI scene suggestions; three editable alternatives, deliberate application and saved ideas.
Director’s desk with ordered shot review, rule-based planning checks, direct navigation and review marks.
Project save/load with thumbnails, browser persistence, eight interface languages and dismissible NEW markers. Some newest monitor/diagnostic labels fall back to English outside HU/EN.
Scope and validation
Windows/Brave tested. Automated suite: 72 passing tests; isolated browser checks cover preview playback, large/narrow layouts, progress, recovery and scoped controls. The author also completed a user-driven H3 shot and confirmed visible progress and automatic saving. These are not universal hardware/workflow/output-quality guarantees.
No models, personal media, credentials or run history are bundled. Optional Ollama/Node.js/ComfyUI and source-workflow dependencies are separate. Runner supports recognized H3 Simple/Pro branches only; no LTX runner, guide-frame inheritance, automatic joining, transition editing or lip-sync.