I created custom nodes for Fish Audio Drama 3 - Works with as little as 8GB VRAM - Quantized weights on HF
Fish Audio Drama 3, converted for the native ComfyUI nodes of ComfyUI-Drama3: a 35B text-to-speech model for voice cloning, dialogue and inline cues such as [sighs] or [whispers], running on cards from 8 GB.
Proof1 community thread
Key points
- 90% of the model is routed experts and each audio frame uses only 8 of the 256 in a layer, so the nodes keep the experts in system RAM and a cache of the ones in use on the GPU.
- Each file holds the whole model in one file; the codec is separate.
- Let me know if yo have any issues.
Sources (1)
- [1]I created custom nodes for Fish Audio Drama 3 - Works with as little as 8GB VRAM - Quantized weights on HFr/StableDiffusion (top, day) · Oct 14, 06:13 PM
Fish Audio Drama 3, converted for the native ComfyUI nodes of ComfyUI-Drama3: a 35B text-to-speech model for voice cloning, dialogue and inline cues such as [sighs] or [whispers], running on cards from 8 GB.
90% of the model is routed experts and each audio frame uses only 8 of the 256 in a layer, so the nodes keep the experts in system RAM and a cache of the ones in use on the GPU.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 14, 2026GPT-Image-style photo editing on your own GPU (12 GB+): I built a zero-setup app for Qwen-Image 2.1 on stable-diffusion.cpp
- Oct 14, 2026mistralai/Timestral-209M
- Oct 14, 2026Train LoRA directly on Quantized GGUFs, no PyTorch, no Parent(FP16) needed, just your GGUF and the data
- Oct 13, 2026Gausify Orbit is now open source: turn AI-generated 360 orbit videos into Gaussian Splats, entirely in your browser
- Oct 13, 2026Run fast, private agents on-device with LFM2.5-24B-A2B
- Oct 13, 2026SonicBoom: 52 tok/s decode on a laptop RTX 3050 (llama.cpp: 45) with the same quantized kernels, the difference is a graph compiler. No Python.
