ByteDance
Seedance 2.0 is the upstream model for clips that benefit most from 4K upscaling — its native output is 1080p in a clean, modern look that holds up to resolution enhancement without revealing artifacts. The pipeline is two-stage: generate the source clip with Seedance 2.0, then route the output through a video upscaler tool node (FAL Topaz Video, FAL Video Upscaler, or your workspace's configured upscaler) for the 4K master. The companion `tools/video-upscale` page covers the upscaler's parameters in detail; this how-to focuses on the Seedance-to-4K pipeline specifically. Default 2x upscale (1080p → 4K equivalent at 3840×2160) is the safe choice for most deliverables; 4x is reserved for hero shots where the longer render and higher credit cost are justified by the use case.
Generate at Seedance 2.0's native settings — don't try to push the output resolution higher inside the generation node. Seedance is optimized for 1080p output; asking for 4K natively gives noisier, less-composed results than a clean 1080p source upscaled in a second step. Same applies to motion: keep the source clip's motion intent simple and well-composed; the upscaler enhances detail but cannot fix bad framing or soft motion. Spend the model's headroom on getting the composition right at 1080p first.
Add a Tool node and select the workspace's configured video upscaler (FAL Topaz Video is the default; alternative routes exist depending on your plan). Connect the Seedance Video node's output to the Tool input. The upscaler runs as an async job through the FAL queue — typical render time is 5-15 minutes for a 5-second clip at 2x, longer for 4x and longer clips. The job tracking lives in the `async_jobs` table and surfaces on the canvas as a progress indicator on the Tool node. While the upscale runs you can keep building other parts of the canvas.
2x is the default for almost everything. 1080p source × 2x = 4K equivalent (3840×2160), which is what most YouTube, broadcast, and high-bitrate social platforms expect. Render time and credit cost are reasonable. 4x (1080p × 4x = 7680×4320, 8K equivalent) doubles the render and roughly doubles the credit cost; reserve it for hero product shots, festival-bound content, or marquee marketing pieces where the extra resolution is visible to the audience. Stacking upscalers (2x → 2x = 4x effective) tends to compound artifacts and is rarely better than a single 4x pass; if you must stack, evaluate the result before committing further credits.
Once the 4K render lands, the canvas timeline lets you export directly to NLE-friendly formats: ProRes 422 HQ for FCPX/Premiere/DaVinci, DNxHR HQX for Avid, or H.264 high-bitrate for direct YouTube/Vimeo upload. The NLE export preserves the timeline structure (any audio nodes, SFX cues, dialogue tracks) as a native sequence rather than a flat render, so a colorist or sound designer can take the project further without losing the canvas structure. For the most common workflow — Seedance source → 2x upscale → ProRes export to Premiere — this entire pipeline runs on Martini without leaving for an external upscaler.
2x is the safe default — 1080p × 2x = 4K equivalent, fits YouTube/broadcast specs, reasonable render time. 4x doubles render and credit cost; use only for hero shots.
Generate at Seedance's native 1080p, don't push native 4K output. A clean 1080p source upscaled produces better results than a noisy native 4K generation.
Stacking 2x → 2x to reach 4x effective compounds artifacts and rarely beats a single 4x pass. If you need 4x, run it once.
The video upscale tool runs as an async job via FAL queue. Render time scales with clip length and upscale factor — budget 5-15 minutes for 5s at 2x.
Companion tool page: `models/tools/video-upscale` covers the upscaler's parameters and routing in detail. This how-to is the Seedance-paired pipeline specifically.
The Seedance 2.0 → video upscale pipeline is the standard 4K master flow on Martini. Native 1080p Seedance output is clean and modern; an upscaler pass takes it to 4K (or 8K for hero shots) without revealing artifacts. The canvas timeline keeps the source clip, upscaled master, audio tracks, and NLE export adjacent, so a video editor can iterate on the cut without leaving the workspace. Trade-off vs. Kling 3.0 paired: Seedance is the cleaner upstream choice for product, lifestyle, and modern social content; Kling 3.0 has the strongest human motion engine, which is the right pick when the source clip is a person performing. Both upscale through the same tool node — the difference is which generator feeds the pipeline.
Connect Seedance 2.0 with other AI models on Martini's infinite canvas. No GPU required — start free.
Get Started Free