SeeDance 2.0 (BudgetPixel)

SeeDance 2.0 is where the Seedance line stops being "animate my image" and becomes a full video suite in a single node. This one node does text-to-video, image-to-video (start plus optional end frame), reference-to-video from up to nine images, and video-to-video editing - you decide the mode by which sockets you connect. That's genuinely convenient, and it's also the reason the credit meter moves fast: 220 credits per second at the top of the range, with 4K resolution available and each mode priced by resolution. It's the most capable Seedance here, and it's the most expensive one too.

The modes, by input

• Text-to-video: prompt only. Length 4–15 seconds, any aspect ratio.
• Image-to-video: wire image (start frame) and optionally end_image; the model interpolates between them.
• Reference-to-video: connect reference_images (up to 9) - note this can't be combined with image/end_image.
• Video editing: connect video and it runs in edit mode, applying the prompt to your clip. Video-edit billing is the sneaky part: you're billed on the input duration at half the output per-second rate plus the output. A 10-second clip isn't a 10-second job.

You can also feed a reference audio track alongside an image or video - but audio can't be the only input.

The inputs that matter

• prompt - required; in video-edit mode it's the edit instruction.
• video - the input clip (2–15s, ≤100MB) for video-to-video. Sets edit mode.
• reference_images - up to 9, reference-to-video.
• length_seconds - 4–15.
• resolution - 480p/720p/1080p/4k. Pricing varies by resolution; 4K is for text/image/reference modes, not video-edit.
• generate_audio - off by default, no price impact when on.
• audio - optional reference audio (MP3/WAV/OGG) that guides the sound.

Output is video (VIDEO) - needs a 2025+ ComfyUI for the core video type.

How it works

Standard pack mechanics: media is serialized and uploaded, a job is created against the API, the node polls (video jobs take minutes) and downloads the finished MP4 into the temp folder. As everywhere in this pack, interrupting the graph abandons the wait, not the job - a submitted generation keeps rendering and billing server-side.

Install and key setup

ComfyUI Manager → BudgetPixel, or:

cd ComfyUI/custom_nodes
git clone https://github.com/budgetpixel-ai/comfyui-budgetpixel.git
pip install -r comfyui-budgetpixel/requirements.txt

Only requests. API key from a paid plan via BUDGETPIXEL_API_KEY or budgetpixel.json; never in workflow JSON.

Where people get burned

Two things. First, the video-edit double bill - input duration at half rate plus output, which most people don't see coming until they hit the cost estimate. Use the BudgetPixel Cost Estimate node before running a video-edit if you care about the number. Second, mode conflicts: image/end_image and video/reference_images are mutually exclusive, so a "wrong-looking" output is usually a stray connection forcing a mode you didn't intend. And remember force_rerun - ComfyUI caches unchanged inputs, so your rerun may be showing you last week's clip.