Add stock MiniMax-H3 video and audio VAEs

NOTICE

@@ -8,3 +8,7 @@ ComfyUI layout, selectively quantized to INT8 ConvRot, and compressed with a
 4,097-row FP32 rank-16 table of the AdaLN time-conditioning curve. It uses the
 standard MiniMax-H3 table path in ComfyUI and does not require modified runtime
 code.
+
+The two safetensors files under vae/ are unmodified, bit-identical copies of
+the Comfy-Org MiniMax-H3 ComfyUI repackages. They retain the MiniMax H3 license
+and are included as standard runtime companions to the modified FL2VA model.

README.md

@@ -32,16 +32,18 @@ standard MiniMax-H3 `adaln_t_table` format and loads in an unmodified ComfyUI.
 | Path | Purpose |
 |---|---|
 | [`FL2VA/minimax-h3-fl2va-int8-lean-convrot-table-k16-g4097-quality21.safetensors`](FL2VA/minimax-h3-fl2va-int8-lean-convrot-table-k16-g4097-quality21.safetensors) | Stock-compatible FL2VA diffusion transformer, 20.940 GiB |
+| [`vae/minimax_h3_video_vae_fp16.safetensors`](vae/minimax_h3_video_vae_fp16.safetensors) | Stock ComfyUI MiniMax-H3 visual VAE, FP16, 4.850 GiB |
+| [`vae/minimax_h3_audio_vae_fp32.safetensors`](vae/minimax_h3_audio_vae_fp32.safetensors) | Stock ComfyUI MiniMax-H3 stereo audio VAE, FP32, 0.564 GiB |
 | [`reports/layer_policy.json`](reports/layer_policy.json) | Exact per-layer precision policy |
 | [`reports/validation.json`](reports/validation.json) | Structural, numerical, and CPU-load results |
 | [`reports/mm_quant_profile_fl2va.json`](reports/mm_quant_profile_fl2va.json) | Row-sampled reconstruction profile for all 200 main matrices |
 
-This repository intentionally does **not** include the MiniMax-H3 text encoder,
-tokenizer, or video/audio VAEs. The text encoder is being prepared as a separate
-repository. Obtain the remaining standard components from the
-[`MiniMaxAI/MiniMax-H3`](https://huggingface.co/MiniMaxAI/MiniMax-H3) or
-[`Comfy-Org/MiniMax-H3`](https://huggingface.co/Comfy-Org/MiniMax-H3)
-repositories. Ref2VA is not included in this release.
+This repository intentionally does **not** include the MiniMax-H3 text encoder.
+It is being prepared as a separate repository. ComfyUI does not require the
+Diffusers scheduler, tokenizer, processor, or remote-code Python bundles as
+separate downloads: the standard workflow needs this diffusion model, the two
+included VAEs, and a compatible MiniMax-H3 text encoder. Ref2VA is not included
+in this release.
 
 ## What was changed
 
@@ -139,9 +141,22 @@ reconstruction result, **not** a perceptual or generation benchmark.
 1. Use a current ComfyUI checkout with MiniMax-H3 and native INT8 ConvRot
    support. This release was CPU-load-tested on a clean, unmodified ComfyUI
    commit `14b05228cef127ce529bc0c08660770d4af3e9a8`.
-2. Place the safetensors file in `ComfyUI/models/diffusion_models/`.
-3. Load it with the standard diffusion-model loader in a MiniMax-H3 FL2VA
-   workflow. Supply the text encoder and video/audio VAEs separately.
+2. Place the FL2VA safetensors file in
+   `ComfyUI/models/diffusion_models/`.
+3. Place both files from `vae/` in `ComfyUI/models/vae/`.
+4. Load the three files with the standard diffusion-model and VAE loaders in a
+   MiniMax-H3 FL2VA workflow. Supply a compatible text encoder separately.
+
+```text
+ComfyUI/models/
+|-- diffusion_models/
+|   `-- minimax-h3-fl2va-int8-lean-convrot-table-k16-g4097-quality21.safetensors
+|-- vae/
+|   |-- minimax_h3_video_vae_fp16.safetensors
+|   `-- minimax_h3_audio_vae_fp32.safetensors
+`-- text_encoders/
+    `-- <MiniMax-H3 text encoder from its separate repository>
+```
 
 No custom node or core patch is required for this checkpoint.
 
@@ -158,6 +173,14 @@ Completed:
   the 4,096 interval midpoints;
 - full CPU load through a clean, unmodified ComfyUI as
   `ModelPatcher -> MiniMaxH3Model`, table mode `(4097, 16)`.
+- both companion VAE files fully loaded on CPU through the same clean ComfyUI:
+  `MiniMaxH3VideoVAE` in FP16 with native internal tiling, and
+  `MiniMaxH3AudioVAE` in FP32 with 32 kHz stereo output;
+- companion VAE SHA-256 checksums match the official Comfy-Org repackaged files:
+  `7c1f131492e7eddacaac9069a61b81bdd39de5cc96561e677c5eab1cdce5e522`
+  (video) and
+  `8e505d95dd1561d47abd43d4238fd40d9bb1ae9e147ed0a4cba778d76ae4db48`
+  (audio).
 
 Not completed at release time:
 
@@ -182,6 +205,17 @@ The official
 was used only as a layout reference and as the source of the FP32
 `rope.inv_freq` tensor absent from the raw Diffusers state dict.
 
+The two files under `vae/` are unmodified, bit-identical copies of the standard
+[`Comfy-Org/MiniMax-H3`](https://huggingface.co/Comfy-Org/MiniMax-H3/tree/main/vae)
+repackages. The original remote-code VAE bundles are more complicated: the
+visual VAE constructs a causal 3D-CNN encoder and 36-layer ViT decoder from
+Python and configuration files, while the audio VAE constructs a DAC-lineage
+encoder and BigVGAN decoder. ComfyUI implements both architectures natively.
+The single-file visual repack casts all 560 source weight tensors from FP32 to
+FP16 and embeds the latent statistics; the audio repack preserves 743 tensors,
+folds 172 weight-normalized pairs into ordinary FP32 weights, and embeds its
+latent statistics. No remote Python code or external VAE config is required.
+
 ## Limitations and intended use
 
 This release is intended for research and local ComfyUI inference. It is a