Fish Audio Drama 3: 35B TTS in ComfyUI running on 8GB VRAM via expert offloading
LatentSpacer · reddit · 2026-10-15
PedroMarinhoDev released ComfyUI-Drama3, native ComfyUI nodes for Fish Audio's 35B text-to-speech model Drama 3, supporting voice cloning, dialogue, and inline cues like [sighs] or [whispers] on cards with as little as 8GB VRAM.
90% of the model is routed experts — each audio frame activates only 8 of 256 experts per layer — so the nodes keep experts in system RAM and cache only active ones on GPU. Quantized weights are on Hugging Face, with example workflows and a demo page included.
Related event: ComfyUI Node Runs 35B Fish Audio Drama 3 TTS Model in 8GB VRAM(2 posts)→
More from Multimodal
- Iris-3B Turbo distills pixel-space diffusion to 4 steps, 1024px image in ~0.7s on one H100 — lmoroney · 2026-10-15
- Stunning AI-generated future racer tearing through a dusk cityscape — ciguleva · 2026-10-15
- Meshy text-to-motion: one sentence description becomes an animation clip — techhalla · 2026-10-15
- Indie dev builds AI game in 4 days: 297 Magnific assets, one-prompt Meshy animations — techhalla · 2026-10-15
- Sundar Pichai raves over Argon demo: pelican on a bike drawn with Fourier epicycles — sundarpichai · 2026-10-15
- Asteria animators build 'The Very Next Day' teaser with custom IC LoRAs on LTX open weights — paultrillo · 2026-10-15