Micron Document
v0.3 latent modality (VAE latents on the wire)

2026-05-09 - v0.3.0 - feature


Image and video diffusion models now stream VAE latents instead of decoded pixels. 48ร— smaller wire weight, decode at the leaf.


Codec v0.3 extends the framing surface from text-tokens to VAE latents. Two new engine forks ship as pre-built Docker images:

โ€ข 'codec-comfyui', ComfyUI with the v0.3 latent transport patch. Production image-gen with the full ComfyUI workflow surface.
โ€ข 'codec-diffusers', the HuggingFace diffusers reference path. Doubles as the bench/golden perceptual-conformance reference for every Codec latent client.

Same wire shape; same registry; same 'LatentStreamDecoder' in '@codecai/web' (https://www.npmjs.com/package/@codecai/web) handles both.

A 512ร—512 RGB frame at fp16 is ~1.5 MB; the SD-1 latent that produced it is 4ร—64ร—64 fp16 = 32 KB, a 48ร— reduction. Per-channel int8 quantization on top, and (for video) delta-coding against keyframes, take it further. The client runs 'vae_decode' locally. Pixels never touch the wire.


Links

GitHub release v0.3.0 (https://github.com/wdunn001/Codec/releases/tag/v0.3.0)

codec-comfyui docs (https://codecai.net/docs/codec-comfyui/)

codec-diffusers docs (https://codecai.net/docs/codec-diffusers/)

PROTOCOL.md ยง Latent Modality (https://github.com/wdunn001/Codec/blob/main/spec/PROTOCOL.md#latent-modality-v03)