Bytedance Seedance 2.0 Image-to-Video

Seedance 2.0 Image-to-Video is the sibling of the pack's Seedance T2V node: instead of a pure text prompt you give it a start frame and describe the motion, and it animates from there. The headline feature is the optional end frame - you can specify both the first and last frames and let the model bridge the motion between them. That's video continuation as a node input, and it's the kind of control you'd normally hack together from multiple generations.

The mechanism is straightforward hosted API: your image goes to WaveSpeed's Seedance 2.0 endpoint as a base64 data URI (if you wire a tensor) or a URL (if you pass a string), the prompt describes the desired motion, and after polling you get a video_url STRING back. It supports the same four model tiers as the T2V node - standard, Turbo, Fast, Fast Turbo - with the same caveat that Turbo variants are 720p/1080p only.

The inputs that matter

• start_frame - an IMAGE tensor, the preferred input; takes precedence over the URL when connected and is sent as base64
• start_frame_url - the fallback string URL when no tensor is connected
• end_frame / end_frame_url - the optional final frame, same tensor-vs-URL pattern; leave both empty for open-ended motion
• prompt - describes the desired motion
• model - the four-tier Seedance dropdown
• duration - 4 to 15 seconds
• aspect_ratio - 16:9, 9:16, 4:3, 3:4, 1:1, 21:9
• resolution - 480p / 720p / 1080p
• generate_audio - native synchronized audio, on by default
• enable_web_search - real-time info during generation
• seed - reproducibility, -1 for random

The tensor-first design is the thing to appreciate here. Because start_frame is a real IMAGE input, you can chain this node straight off anything that outputs a tensor - a local checkpoint's VAE decode, an edit node, even another video's first frame extracted to a tensor. No hosting a URL, no Load Image round-trip.

Install and setup

cd ComfyUI/custom_nodes
git clone https://github.com/eRepublik-Labs/comfyui-nodes-erpk.git erpk
cd erpk && pip install -r requirements.txt

Or "ERPK Custom Nodes" from ComfyUI Manager, restart, and set a WaveSpeed API key in ERPK Settings (right-click canvas). Output goes to Preview Anything like every other video node here.

Troubleshooting

If you've read the T2V article, the gotchas carry over: Turbo tiers reject 480p, generate_audio bills by default, and reference/duration caps are enforced server-side. The one node-specific mistake is wiring both start_frame and start_frame_url with conflicting content - the tensor silently wins, so if you're fighting "why is it using the wrong image," check whether a tensor is connected. Same for end_frame. And a practical tip: for a clean loop or a seamless A→B transition, put real thought into the end frame's composition - the model bridges what's there, and a wildly different end frame gives you a wild jump.