PiD Decoder: Generate at One Megapixel, Decode Straight to 4K
Four ComfyUI graphs using the PiD pixel diffusion decoder to turn a ~1 MP latent into 2K/4K output without a separate upscaler.
Contents
README.md- this tutorial
What ships
- PiD node chain (
PiDPrepare/PiDSample/PiDFinalizex/PiDKSamplerCapture) in four base-model graphs. - FLUX.1 Dev 28 steps CFG 1; FLUX.2 Dev 50 steps CFG 4; SD3.5 Large 25 steps CFG 4.4; Z Image Turbo 9 steps CFG 1.
- ~16x decode factor, so a 1 MP first pass lands near 4K.
Deployment
- Import a graph and install the PiD custom nodes.
- Generate at ~1 MP, then let PiD decode; wire PiD sigma directly unless you need manual sigma.
- Prefer a rented 48 GB instance; local runs are OOM-prone per the author.
Links
- RunningHub - FLUX.1 Dev PiD 4K: https://www.runninghub.ai/post/2059961777689030657/?inviteCode=rh-v1111
- RunningHub - FLUX.2 Dev PiD 4K: https://www.runninghub.ai/post/2059961769686294530/?inviteCode=rh-v1111
- RunningHub - SD3.5 Large PiD 4K: https://www.runninghub.ai/post/2059961785033256961/?inviteCode=rh-v1111
- RunningHub - Z Image Turbo PiD 4K: https://www.runninghub.ai/post/2059961793564471298/?inviteCode=rh-v1111
- YouTube: https://www.youtube.com/watch?v=nPt4IJ03Y9Q
- Bilibili: https://www.bilibili.com/video/BV1tsVH6kEBJ
Limits
- PiD enlarges defects instead of fixing them (SD3.5 hands are the example).
- High sigma degrades output.
- VRAM hungry; 48 GB can still fail.
Optional API platform - 88API (affiliate link)
Explore 88API for access to multiple AI model APIs. Available models, routes and prices are listed on the platform; Alipay top-ups are supported.
https://88api.ai/sign-up?aff=usx3
Disclosure: I may earn a commission from eligible purchases after you register through this referral link.
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐ Ask for provider support