EvoLink Veo 3.1 Image-to-Video (Official)

The image-to-video node is where Veo 3.1 actually shines. Text-to-video is fine, but pointing a real frame at the model and asking it to move is how you keep a composition, a character, a mood you already made elsewhere in the graph. EvoLink Veo 3.1 Image-to-Video does exactly that - and unlike most of the pack, first_frame is a required input, because without a first frame this node has no reason to exist.

How it works

Standard EvoLink plumbing: submit to /videos/generations, poll, download before the 24-hour result links expire, return a native VIDEO output. The first_frame image uploads through EvoLink's own file channel - no third-party host, no pasting URLs - and you can optionally add a last_frame so the clip ends on a specific image rather than wherever the model drifts. First-frame-only is the casual setup; first + last frame is a closed loop and honestly the more impressive demo.

Model choices are veo-3.1-generate-preview and veo-3.1-fast-generate-preview. aspect_ratio defaults to auto, which just inherits your first frame's proportions - leave it alone unless you want to force a crop. Quality goes up to 4k, and like its text-to-video sibling, audio generation defaults to on and costs extra. Turn generate_audio off while iterating.

The inputs that matter

• first_frame - required IMAGE input. Load Image → here.
• last_frame - optional IMAGE input for a fixed ending.
• prompt - required, ~2000 tokens, Chinese-friendly; describes the motion and scene.
• model - full vs. fast Veo 3.1.
• quality - 720p / 1080p / 4k.
• generate_audio - native audio, extra charge, on by default.
• seed / negative_prompt - for reproducible takes and steering away from things.

Outputs

video → Save Video. Plus the shared EvoLink quartet: task_id (job-log receipt), status_info (model/time/credits), result_urls (24h raw links), response_json (full payload). Wire result_urls into the Topaz upscale node for a generate-then-sharpen chain.

Install and gotchas

Part of the comfyui-evolink pack. Manager → search EvoLink → Install → fully restart ComfyUI (close the console, not just refresh). Or:

cd ComfyUI/custom_nodes
git clone https://github.com/deeplearning-goethe/comfyui-evolink

Only requests as a dependency; no models, no GPU. Key from evolink.ai/dashboard/keys, fill api_key once, blank afterward so it doesn't leak into shared JSON. EVOLINK_API_KEY env var works.

The one gotcha that's specific to this node: it's the rare EvoLink node where an input is genuinely required, so if you hit "prompt/first_frame can't be empty"-style errors, check you've actually wired a Load Image into first_frame. Everything else follows pack convention - 401 bad key, 402 no credit, 429 rate limit, moderation blocks free, unchanged reruns cached and unbilled. And if you want subject/style consistency across a whole clip rather than a fixed first/last frame, the sibling EvoLink Veo 3.1 Reference node is the one.