EvoLink Kling V3 Image-to-Video (Official)

Kling V3 is the model that made Kuaishou's name in the AI-video race, and EvoLink_KlingV3I2V is its image-to-video node in this pack. This is the full-featured workhorse: up to 4K, a negative prompt, sound on tap, and both a first and optional last frame so you can pin down where the clip starts and where it ends. If you're animating from a strong image and want the finished-feeling result - not the budget-tier draft - this is the Kling node you'll actually keep in your main workflow.

The inputs that matter

• first_frame (required) - the IMAGE input that starts the clip. Must be at least 300px per side with a width:height ratio between 1:2.5 and 2.5:1. Square-ish frames are safest.
• prompt (required) - shot and motion description, up to 2500 characters, Chinese-friendly.
• last_frame (optional) - a second IMAGE input. Connect it for the first/last-frame trick: the model animates from one to the other, which gives you real control over the ending instead of hoping.
• duration - 3–15 seconds, default 5, billed by duration.
• quality - 720p (default), 1080p, or 4k.
• sound - off (default) or on; generated sound effects.
• negative_prompt - things you don't want in the clip. This is the field Turbo drops, so if you need it, this is the right node.

Then the pack-standard api_key and timeout_seconds (1200).

How it works

Your frames upload through EvoLink's file channel (auto-deleted after 72h, no third-party host), the job posts to POST /v1/videos/generations as kling-v3-image-to-video, and the pack polls until the clip completes, downloading it as a VIDEO object. Connect video to a native Save Video node. The other four outputs - task_id, status_info, result_urls (24h expiry), response_json - are the pack-wide standard.

Installing

No GPU, no model files, only requests:

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

then fully restart ComfyUI, or search EvoLink in ComfyUI Manager. If nodes are missing, check the folder isn't nested two levels deep.

Troubleshooting

401 key, 402 balance, 403 access, 429 rate limit, plus the VPN note - whitelist api.evolink.ai and files-api.evolink.ai. Kling-specific traps: keep the first frame within the 300px minimum and the 1:2.5–2.5:1 ratio, or the API rejects it with a message that won't always translate to "resize your image." Watch the cost curve - 4K at 15 seconds is a different beast than 720p at 5 seconds, so draft low and final high. And if sound makes the job slow, that's normal; audio generation rides along with the video pass.