Edit Video

What it is

DeepGen_V2V0 is the video-to-video editor in the DeepGen pack. You feed it an existing clip plus an instruction prompt, and it returns a new version of that clip that follows your text - a restyle, a scene change, a fixed piece of motion, whatever survives the model's interpretation. The display name, "Edit Video," undersells how much of the video frontier this sits on: as of the current pack version, the model dropdown has exactly one entry, Grok Imagine Video (xAI's video model). It's a one-model shop today, and that's worth knowing before you build a whole workflow on it.

How it works

The mechanism is shared with the pack's other video nodes. The input clip is base64-encoded and posted to the API as an attachment alongside your prompt. The job goes into a queue, the node polls every 15 seconds until it's done, then downloads the result into ComfyUI's temp folder as the pack's VIDEO type. "Edit" is doing a lot of work server-side - you send a video up, you get a video back, and the model re-renders with your prompt as the control signal.

Inputs and outputs

• model - Grok Imagine Video, and that's the whole list right now.
• prompt - the edit instruction. "Make it nighttime," "change the character's jacket to red," "turn this into a slow-motion cinematic shot." The prompt is 100% of the steering, so this is where your time goes.
• seed_value, nb_results (1–10, but as with the other video nodes only the first result returns while the rest bill you - keep it at 1), output_prefix.
• config_json - extra JSON params for the API; blank is fine.
• Optional: video, aspect_ratio, resolution (Grok supports 480p and 720p per the pack's model table), duration (1–10s for this model), generate_audio (on by default).

Outputs: VIDEO (the edited clip), output_prefix_and_model (STRING, handy for naming), total_credits_used (FLOAT - your meter).

Wiring

The video input takes the pack's VIDEO type, so the obvious sources are a DeepGen_T2V0 result or the DeepGen_LVID Load Video node. Generate → edit → review → edit again is the loop, and it's why all these nodes ship in one pack.

Install and key

Pack standard: ComfyUI Manager → search "ComfyUI-DeepGen-API" → Install, or clone https://github.com/deepiksdev/ComfyUI-DeepGen-API.git into custom_nodes and pip install -r requirements.txt (just requests and opencv-python). Restart, then put your key in via the Settings gear (DeepGen API Key, from deepgen.app) - it's stored in ComfyUI/user/deepgen/config.json. No model files to download; the model runs on DeepGen's servers.

Honest caveats

The single-model lock-in is the big one. If xAI's Grok service changes or the pack swaps models, this node is whatever the dropdown says - there's no local fallback baked in. Every edit costs credits, and video edits cost more than text calls. And your clip leaves the machine: with V2V you're not just sending a prompt, you're shipping the entire source video to a server you don't control, with its moderation filters applied at the source. If that's a dealbreaker, this isn't the node for the job - but for quality text-guided video editing from inside ComfyUI, there isn't much else at this level.