VRGDG Video Editor UI

The VRGDG Video Editor UI is the hub of the pack's music-video pipeline. Where the Remake Clip Queue and Session Loader operate on single clips, this is the node you open to see and edit the whole project: browse rendered clips, pick one, capture a frame, generate scene prompts from it, flag it for remake, and save the session that every other node reads. It's the bridge between "the AI did its thing" and "now let me make it good."

You point it at a project's output folder, and it takes over from there. Everything it touches ends up in a session JSON (vrgdg_temp/editor_session.json) that the queue and the session loader consume - so the editor is where your decisions become data.

The inputs

• output_folder - the project output folder to load. This is your clip library.
• video_extensions (default .mp4,.mov,.webm,.mkv) - comma-separated extensions the editor shows. If your clips are hiding, they're probably an extension not on the list.
• selected_clip_path / captured_frame_path - the currently selected clip and the last frame captured from it. These update as you work.
• session_path - where the session JSON lives.
• model_file / mmproj_file / i2v_model_file - the Gemma GGUF models for the editor's visual prompting. The defaults read [No Gemma GGUF found in models/LLM] - that placeholder is the pack telling you it couldn't find a model, not a model name. Drop GGUFs into models/LLM and the enums populate. The visual prompt button needs both a Gemma GGUF and its mmproj (so the model can see the captured frame); the I2V prompt button only needs a text-only Gemma GGUF.
• generated_t2i_prompt / generated_i2v_prompt - the last prompts the editor's Gemma buttons produced.

The outputs

The same state, passed onward so you can wire it: output_folder, session_path, captured_frame_path, generated_t2i_prompt, and generated_i2v_prompt. That's how the editor hands the generated prompts into your sampler graph without you copying text.

How it works

The workflow loop is: load the folder, review clips, capture a frame from the one you're unhappy with, hit the visual prompt button (Gemma + mmproj reads the frame and writes a scene prompt), tweak it, mark it for remake, save. Downstream, the Remake Clip Queue walks the saved session and re-renders just the flagged clips. The editor makes the "review → improve → re-render" cycle feel deliberate instead of chaotic.

Install and the real gotchas

Pack install is standard (Manager → search vrgamedev, restart, hard-refresh - this is the most JS-heavy node in the pack, so the hard refresh is not optional). The honest friction is the Gemma side:

• Put the GGUF files in models/LLM and restart so the file lists refresh. If the enums still show [No Gemma GGUF found...], the pack can't see them.
• llama-cpp-python is the underlying loader and is the pack's fiddliest dependency on Windows - it needs Cython + scikit-build-core (README covers this) and sometimes a working C++ toolchain; Python 3.12 is the safer environment. If the prompt buttons silently do nothing, this is where to look.

Also worth knowing: the author's community walkthroughs note the UI changes fast - features land "almost daily" per the beta announcements, so videos and screenshots from last month may not match what you see. That's a sign the pack is alive, but keep the README's Video Builder guide as your current source of truth.