Seedance 1.0 Pro

Seedance is ByteDance's text-to-video line, and 1.0 Pro is the version that earned the reputation: strong motion, coherent scenes, and - the thing people actually notice - camera control that doesn't feel like an afterthought. In this pack it's a single node under Runware/Video/bytedance: type a prompt, optionally hand it a starting frame or two, and a native VIDEO comes back from Runware's cloud. No local GPU, no weights, no ffmpeg dance to get a result - you pay per run and it renders server-side.

How it works

The node sends your prompt (and any frameImages) to Runware, ByteDance's Seedance 1.0 Pro generates the clip, and the result is decoded into a native VIDEO on your machine. It's text-to-video out of the box, but the frameImages input is what makes it image-to-video too - feed it a still and the model animates from that frame instead of inventing one. The cameraFixed option is the quiet killer feature: lock the camera and you get a stable shot rather than the roaming, drift-prone camera movement that made early text-to-video feel cheap.

The inputs that matter

• positivePrompt (required, multiline) - what you want on screen. This is a video model, so write for motion and scene, not just subject: lighting, camera behavior, and action read better than a list of tags.
• frameImages (IMAGE) - one or more start frames for image-to-video. Skip it for pure text-to-video.
• size - a dropdown of 17 presets: None (model default), plus 480p/1080p in common aspect ratios (16:9, 4:3, 1:1, 3:4, 9:16, 9:21, 21:9). Only one size is sent, so pick the preset and don't fight it.
• duration (1.2–12 seconds) - clip length. The tooltip notes total frames = duration × the model's fps. Longer isn't automatically better; 5 seconds is the default for a reason.
• seed - reproducibility. Set it to lock a take.
• numberResults (1–4) - seed-based variations. For video, where a bad run costs real credits, generating 2–3 and picking the winner often beats rerolling.
• providerSettings.bytedance.cameraFixed (BOOLEAN) - locks camera position and angle, preventing pan/tilt/zoom. On for anything where the scene should hold still.
• safety (toggle gating safety.checkContent and safety.mode none/fast/full) - leave the content check on for shared workflows; the node prints the verdict in its title bar.
• outputFormat (MP4/WEBM/MOV) and outputQuality (20–99) - delivery and compression.

Wiring it up

Output is video → any VIDEO preview/save node. For image-to-video, pull frameImages from a generation node or a loaded image. You can chain reference videos via the RunwareBuild_referenceVideos builder if the model exposes the socket, but Seedance 1.0 Pro's strength is single-prompt and single-frame control - keep it simple.

Install and setup

One install for the whole catalog: ComfyUI Manager → "Runware" → install → restart, or clone and pip install -r ComfyUI-Runware/requirements.txt (runware-sdk, pillow, soundfile). API key in Settings → Runware API key, RUNWARE_API_KEY, or runware auth login, plus funded credits. Video inference is the priciest thing in this pack, and the community's recurring friction is the ~$20 minimum top-up with a card on file - budget before you batch.

Troubleshooting

• Video lands as a file path - older ComfyUI lacks VideoFromFile, so the pack saves to your output folder instead of wrapping a VIDEO. Check there.
• Camera wanders when you wanted a lock - cameraFixed exists for exactly this. Flip it on.
• Clips come out short/choppy - raise duration toward the model's comfort zone and lower outputQuality if you're iterating; re-encode at high quality only for the final.
• Blocked by safety - safety.mode: full can reject borderline content. Adjust content or drop to none if your material is clearly fine.

Seedance 1.0 Pro is the "just make me a clip" node of this catalog - good motion, controllable camera, and a single obvious place to plug in a starting frame. It won't replace a real production pipeline, but for iterating on a video idea it's the most direct path from prompt to playback.