Can I generateimages, video, voices, musicon this machine?
We read your GPU through your browser, then check it against every local generative model worth running — FLUX, SDXL, Wan 2.2, HunyuanVideo, LTX-Video, Qwen-Image, plus Kokoro, XTTS, Bark, Orpheus for speech and ACE-Step, MusicGen, YuE, Stable Audio for music. No sign-ups. No uploads. Just an answer.
Software to actually run these models
Each one runs locally on Linux— pick a frontend, point it at the model files, and you’re generating.
Node-based pipeline editor — the de-facto standard for local diffusion. Supports every image and video model in this catalog.
One-click installer for hundreds of local AI apps — ComfyUI, Wan2GP, AudioCraft, Bark, F5-TTS, ACE-Step, all in a single launcher.
User-friendly frontend that runs ComfyUI under the hood. Same model coverage, much friendlier than building graphs.
Memory-optimised Wan 2.1 / 2.2 + Hunyuan + LTX video runner for the GPU-poor. Ships with a Gradio UI.
All-in-one Gradio for Bark, Tortoise, MMS, MusicGen, AudioGen, RVC and more. Pinokio-installable.
Stable Cascade
3-stage cascade (Würstchen architecture). Strong prompt adherence, lighter than its 5.9B suggests.
Z-Image Turbo
Distilled few-step model. FP16 fits comfortably in 16GB. Apache-licensed.
Z-Image Base
Full (non-distilled) Z-Image. More steps than Z-Image Turbo but higher fidelity. Apache-licensed.
Qwen-Image-2.1
Auto-added from Hugging Face (Qwen/Qwen-Image-2.1). VRAM/quant/disk figures are estimated from parameter count — review and refine.
qwen2-audio-7b-instruct-gguf
Auto-added from Hugging Face (forkjoin-ai/qwen2-audio-7b-instruct-gguf). VRAM/quant/disk figures are estimated from parameter count — review and refine.
Stable Diffusion XL
The workhorse. Sharp 1024px output, huge fine-tune ecosystem (Pony, Illustrious, Juggernaut).
SDXL Turbo
1-step distilled SDXL. Generates in under a second on midrange GPUs. Lower fidelity than full SDXL.
OmniGen 2
Unified text-to-image and image-editing model. One model handles generate, edit, compose.
CogVideoX-5B
5B T2V/I2V from Tsinghua. Mid-range hardware target. 6-second clips at 720p.
FLUX.2 [klein] 4B
The smaller Apache-licensed FLUX.2. 4B params at 1024px — fits comfortably on midrange GPUs and high-end phones.
FLUX.2 Klein
4 B distilled from FLUX.2 32 B. Real-time generation on consumer GPUs at FLUX-grade fidelity.
ACE-Step 1.5
Best open full-song generator with vocals. Generates 4-minute tracks in ~20 s on a 4090.
MusicGen Large
Meta's flagship MusicGen. Strong instrumental tracks. Non-commercial only.
MusicGen Stereo Large
Stereo variant of MusicGen Large. True L/R channels for richer mix.
s2-pro
Auto-added from Hugging Face (fishaudio/s2-pro). VRAM/quant/disk figures are estimated from parameter count — review and refine.
YuE2-3B
Auto-added from Hugging Face (m-a-p/YuE2-3B). VRAM/quant/disk figures are estimated from parameter count — review and refine.
Voxtral-4B-TTS-2603
Auto-added from Hugging Face (mistralai/Voxtral-4B-TTS-2603). VRAM/quant/disk figures are estimated from parameter count — review and refine.
Breeze-TTS-2
Auto-added from Hugging Face (BreezeBlue/Breeze-TTS-2). VRAM/quant/disk figures are estimated from parameter count — review and refine.
higgs-tts-3-4b
Auto-added from Hugging Face (bosonai/higgs-tts-3-4b). VRAM/quant/disk figures are estimated from parameter count — review and refine.
ideogram-4-nf4
Auto-added from Hugging Face (ideogram-ai/ideogram-4-nf4). VRAM/quant/disk figures are estimated from parameter count — review and refine.
acestep-v15-xl-turbo
Auto-added from Hugging Face (ACE-Step/acestep-v15-xl-turbo). VRAM/quant/disk figures are estimated from parameter count — review and refine.
Wan2.2-TI2V-5B-GGUF
Auto-added from Hugging Face (QuantStack/Wan2.2-TI2V-5B-GGUF). VRAM/quant/disk figures are estimated from parameter count — review and refine.
FLUX.2-klein-base-4B
Auto-added from Hugging Face (black-forest-labs/FLUX.2-klein-base-4B). VRAM/quant/disk figures are estimated from parameter count — review and refine.
Wan2.2-TI2V-5B-Diffusers
Auto-added from Hugging Face (Wan-AI/Wan2.2-TI2V-5B-Diffusers). VRAM/quant/disk figures are estimated from parameter count — review and refine.
acestep-v15-xl-base
Auto-added from Hugging Face (ACE-Step/acestep-v15-xl-base). VRAM/quant/disk figures are estimated from parameter count — review and refine.
acestep-v15-xl-sft
Auto-added from Hugging Face (ACE-Step/acestep-v15-xl-sft). VRAM/quant/disk figures are estimated from parameter count — review and refine.
FLUX.2-klein-4b-fp8
Auto-added from Hugging Face (black-forest-labs/FLUX.2-klein-4b-fp8). VRAM/quant/disk figures are estimated from parameter count — review and refine.
MOSS-TTS-Local-Transformer-v1.5
Auto-added from Hugging Face (OpenMOSS-Team/MOSS-TTS-Local-Transformer-v1.5). VRAM/quant/disk figures are estimated from parameter count — review and refine.
HeartMuLa-oss-3B-happy-new-year
Auto-added from Hugging Face (HeartMuLa/HeartMuLa-oss-3B-happy-new-year). VRAM/quant/disk figures are estimated from parameter count — review and refine.
DreamX-World-5B
Auto-added from Hugging Face (GD-ML/DreamX-World-5B). VRAM/quant/disk figures are estimated from parameter count — review and refine.
SeFi-Image-5B-turbo
Auto-added from Hugging Face (SeFi-Image/SeFi-Image-5B-turbo). VRAM/quant/disk figures are estimated from parameter count — review and refine.
YuE2-3B
Auto-added from Hugging Face (mrfakename/YuE2-3B). VRAM/quant/disk figures are estimated from parameter count — review and refine.
Stable Diffusion 3 Medium
MMDiT architecture with triple text encoders. Better text rendering than SDXL.
Stable Diffusion 3.5 Medium
Improved 3.5 generation with better composition and aesthetics than SD3 Medium.
Lumina Image 2.0
Compact next-gen DiT with strong photographic realism. Apache-licensed for commercial use.
LTX-Video
Realtime-capable video model. Generates faster than playback on a 4090. The speed champion.
Kolors
Strong photographic realism, multilingual prompts (Chinese + English). Mid-weight footprint.
Parler-TTS Large
Larger Parler with broader voice and prosody range. Slower, higher fidelity.
Orpheus 3B
LLM-style TTS with rich emotional control. Streaming output via vLLM.
VibeVoice-1.5B
Auto-added from Hugging Face (microsoft/VibeVoice-1.5B). VRAM/quant/disk figures are estimated from parameter count — review and refine.
Qwen3-TTS-12Hz-1.7B-CustomVoice
Auto-added from Hugging Face (Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice). VRAM/quant/disk figures are estimated from parameter count — review and refine.
VoxCPM2
Auto-added from Hugging Face (openbmb/VoxCPM2). VRAM/quant/disk figures are estimated from parameter count — review and refine.
MiniMax-Music3
Auto-added from Hugging Face (MiniMaxAI/MiniMax-Music3). VRAM/quant/disk figures are estimated from parameter count — review and refine.
Qwen3-TTS-12Hz-1.7B-VoiceDesign
Auto-added from Hugging Face (Qwen/Qwen3-TTS-12Hz-1.7B-VoiceDesign). VRAM/quant/disk figures are estimated from parameter count — review and refine.
stable-audio-3-medium
Auto-added from Hugging Face (stabilityai/stable-audio-3-medium). VRAM/quant/disk figures are estimated from parameter count — review and refine.
dots.tts-soar
Auto-added from Hugging Face (rednote-hilab/dots.tts-soar). VRAM/quant/disk figures are estimated from parameter count — review and refine.
Yue2-3B-GGUF
Auto-added from Hugging Face (audio-cpp/Yue2-3B-GGUF). VRAM/quant/disk figures are estimated from parameter count — review and refine.
Kizagan-TTS-v1.0
Auto-added from Hugging Face (AlicanKiraz0/Kizagan-TTS-v1.0). VRAM/quant/disk figures are estimated from parameter count — review and refine.
midashenglm-gen
Auto-added from Hugging Face (mispeech/midashenglm-gen). VRAM/quant/disk figures are estimated from parameter count — review and refine.
stable-audio-3-medium-base
Auto-added from Hugging Face (stabilityai/stable-audio-3-medium-base). VRAM/quant/disk figures are estimated from parameter count — review and refine.
magenta-realtime-2
Auto-added from Hugging Face (magenta-community/magenta-realtime-2). VRAM/quant/disk figures are estimated from parameter count — review and refine.
YuE2-3B-MLX
Auto-added from Hugging Face (ahmadw/YuE2-3B-MLX). VRAM/quant/disk figures are estimated from parameter count — review and refine.
LTX-Video-2B-CoreAI
Auto-added from Hugging Face (mlboydaisuke/LTX-Video-2B-CoreAI). VRAM/quant/disk figures are estimated from parameter count — review and refine.
Anima
Anime-focused 2B text-to-image model from CircleStone Labs using NVIDIA Cosmos-Predict2-2B-Text2Image architecture, optimized for anime and illustration rather than realism.
IndexTTS-2.5
Zero-shot text-to-speech model that clones a voice from a single reference clip. It supports Chinese, English, Japanese, Spanish, and Arabic with cross-lingual voice transfer and emotion control.
Hunyuan-DiT
Bilingual (Chinese + English) DiT. Strong at Chinese text rendering, light to run.
Stable Video Diffusion
Image-to-video, ~14–25 frames at 576×1024. The original consumer-friendly local video model.
Wan 2.1 T2V 1.3B
Tiny T2V. The most compatible video model — fits on 8GB GPUs. Apache-licensed.
SkyReels V2 1.3B
Lightweight successor with infinite-length generation. Fits on 8 GB GPUs.
SANA 1.5
Linear DiT that generates up to 4K images on as little as 8 GB. Tiny weights, massive output.
Bark
Generates speech, laughter, sighs, music — all from a text prompt. Quirky but expressive.
Parler-TTS Mini
Prompt-controlled TTS — describe the voice in natural language ('young woman, warm, slow').
Sesame CSM 1B
Conversational speech model — context-aware prosody for dialogue agents.
Fish Speech 1.5
Multilingual zero-shot voice cloning. Strong on Chinese, Japanese, Korean.
MusicGen Medium
Mid-tier MusicGen. Faster than Large, slightly less rich.
Stable Audio Open 1.0
Sound effects, ambient textures, short instrumental loops up to 47 s.
Stable Audio Open 1.5
Refined sound design model. Cleaner ambient textures, tighter SFX.
DiffRhythm
Latent-diffusion full-song generator. End-to-end. 8 GB minimum with chunked inference.
Qwen3-TTS-12Hz-0.6B-Base
Auto-added from Hugging Face (Qwen/Qwen3-TTS-12Hz-0.6B-Base). VRAM/quant/disk figures are estimated from parameter count — review and refine.
Qwen3-TTS-12Hz-0.6B-CustomVoice
Auto-added from Hugging Face (Qwen/Qwen3-TTS-12Hz-0.6B-CustomVoice). VRAM/quant/disk figures are estimated from parameter count — review and refine.
clark-air-sana-1.6b-1.58bit
Auto-added from Hugging Face (clark-labs/clark-air-sana-1.6b-1.58bit). VRAM/quant/disk figures are estimated from parameter count — review and refine.
FastWan-QAD-FP8-1.3B
Auto-added from Hugging Face (FastVideo/FastWan-QAD-FP8-1.3B). VRAM/quant/disk figures are estimated from parameter count — review and refine.
sanoTTS
Tiny neural text-to-speech voices from Ampixa. The repo packages multiple standalone voices and supports browser, Python, and Arduino-style local inference.
Confucius4-TTS
Multilingual LLM-based text-to-speech system with zero-shot cross-lingual voice cloning and a learnable speaker encoder.
Chatterbox
Production-grade open-source text-to-speech model family with English and multilingual variants, supporting zero-shot voice cloning and emotion control.
Stable Diffusion 1.5
The classic. Tiny, fast, runs on almost anything. Massive ecosystem of LoRAs and fine-tunes.
Stable Diffusion 2.1
Successor to SD 1.5 with native 768px training. Smaller community than 1.5 but still light on hardware.
PixArt-Σ
Featherweight 0.6B DiT with T5-XXL encoder. Beautiful 4K output for the param count.
Chatterbox Turbo
350 M model with a 1-step distilled decoder. Sub-200 ms latency for production agents.
F5-TTS
Flow-matching TTS with strong voice cloning from a 6-second reference clip.
XTTS v2
Multi-language voice cloning from 6 seconds of reference audio. 17 languages.
Spark-TTS 0.5B
Tiny LLM-driven zero-shot voice cloning. 4 GB VRAM, runs comfortably on most GPUs.
Kani-TTS 2
400 M streaming TTS on a Liquid LFM2 backbone with NVIDIA NanoCodec. 3 GB VRAM end-to-end.
MusicGen Small
Tiny MusicGen for fast prototyping. Runs on a 4 GB GPU.
Magenta RealTime
Live, prompt-steerable instrumental generation. Optimized for streaming.
OmniVoice
Auto-added from Hugging Face (k2-fsa/OmniVoice). VRAM/quant/disk figures are estimated from parameter count — review and refine.
Irodori-TTS-v4.1-Anime
Auto-added from Hugging Face (phasefield-audio/Irodori-TTS-v4.1-Anime). VRAM/quant/disk figures are estimated from parameter count — review and refine.
Irodori-TTS-500M-v3
Auto-added from Hugging Face (Aratako/Irodori-TTS-500M-v3). VRAM/quant/disk figures are estimated from parameter count — review and refine.
stable-audio-3-small-music
Auto-added from Hugging Face (stabilityai/stable-audio-3-small-music). VRAM/quant/disk figures are estimated from parameter count — review and refine.
sarashina2.2-tts
Auto-added from Hugging Face (sbintuitions/sarashina2.2-tts). VRAM/quant/disk figures are estimated from parameter count — review and refine.
stable-audio-3-small-sfx
Auto-added from Hugging Face (stabilityai/stable-audio-3-small-sfx). VRAM/quant/disk figures are estimated from parameter count — review and refine.
stable-audio-3-small-music-base
Auto-added from Hugging Face (stabilityai/stable-audio-3-small-music-base). VRAM/quant/disk figures are estimated from parameter count — review and refine.
stable-audio-3-small-sfx-base
Auto-added from Hugging Face (stabilityai/stable-audio-3-small-sfx-base). VRAM/quant/disk figures are estimated from parameter count — review and refine.
magenta-realtime-2-small
Auto-added from Hugging Face (magenta-community/magenta-realtime-2-small). VRAM/quant/disk figures are estimated from parameter count — review and refine.
IndicF5
IndicF5 is a near-human polyglot text-to-speech model for 11 Indian languages. It generates speech from a reference prompt audio and matching text.
Stable Diffusion 3.5 Large
Stability's flagship 8B MMDiT. Sharp text, strong composition. Triple text encoder (CLIP-L, CLIP-G, T5-XXL).
AuraFlow v0.3
Truly open-source flow-matching model. The largest fully Apache-licensed image model.
FLUX.1 schnell
4-step distilled FLUX. The fastest way to get FLUX-tier quality. Apache-licensed.
FLUX.1 dev
The model to beat. State-of-the-art prompt adherence and photorealism. Hungry but worth it.
FLUX.1 Kontext dev
Image editing variant of FLUX. Reference image + prompt → edited result.
HiDream-I1
Hybrid DiT + MoE. Beats Flux on several benchmarks. MIT-licensed for any use.
Qwen-Image
20B MMDiT from Alibaba. Best-in-class text rendering — handles paragraphs of text in images.
Qwen-Image-Edit
Edit variant of Qwen-Image. Replace, add, restyle objects via natural prompts.
Hunyuan Image 2.1
Tencent's 17B image flagship. Strong at composition and Chinese text.
Hunyuan Image 3
Massive 84B MoE image model with 13B active. Quality rivals closed-source flagships. Needs serious hardware.
FLUX.2 dev
Successor to FLUX.1 dev. Higher fidelity, longer context. Heavy hardware required.
ERNIE-Image
Baidu's 10B image model. Strong at Chinese prompts and stylized output.
HiDream-E1.1
Editing variant of HiDream-I1. Same MoE backbone tuned for image editing.
Wan 2.1 T2V 14B
Full-size Wan T2V. Strong motion, 720p output. Quantization makes it viable down to 12GB.
Wan 2.1 I2V 14B
Image-to-video flagship. Best open I2V motion in early 2025.
Wan 2.2 TI2V 5B
Unified text + image to video. Best-in-class T2V/I2V at 5B. Fits on a 24GB GPU at FP16.
Wan 2.2 T2V A14B (MoE)
MoE flagship — 27B total / 14B active. Top open-video quality.
LTX-2
Successor to LTX-Video. 4K-capable on high-end hardware. Distilled variants run on 12GB.
HunyuanVideo
13B T2V — closest open rival to Sora at release. Slow but cinematic.
HunyuanVideo 1.5
Lighter, sharper successor. 8.3B params, 14GB minimum, runs on consumer GPUs.
Mochi 1
10B asymmetric DiT. Ambitious motion. Apache-licensed for any use.
Pyramid Flow
Pyramidal flow-matching for efficient long videos. Strong T2V quality at 2B.
Chroma 1 HD
Community 8.9B distillation of FLUX.1-schnell, fully Apache 2.0. Strong photographic quality with no commercial restrictions.
FLUX.1 [klein]
Apache-licensed 9B distillation of FLUX.1. Smaller and lighter than schnell while keeping FLUX-tier prompt adherence.
FLUX.2 [klein] 9B
The larger Apache-licensed FLUX.2. 9B params, sharper detail than the 4B variant.
SkyReels V1
Human-centric video foundation model based on HunyuanVideo. Up to 12 s clips at 24 fps, 544×960. Strong at faces and motion.
SkyReels V2 14B
Flagship infinite-length video model derived from Wan 2.1 14B. Top-tier open-source motion at 720p.
Cosmos-Predict2 2B Text2Image
Physics-aware text-to-image — coherent geometry and lighting, tuned for sim/robotics.
Cosmos-Predict2 2B Video2World
World-model T2V/I2V. Generates physics-consistent motion. Heavy VRAM appetite.
Kokoro 82M
Featherweight TTS that punches way above its weight. Sub-300 ms inference, runs on CPU.
YuE 7B
Suno-style full-song generation with synchronized lyrics + vocals. Up to 5-minute tracks.
WAN2.2-14B-Rapid-AllInOne
Auto-added from Hugging Face (Phr00t/WAN2.2-14B-Rapid-AllInOne). VRAM/quant/disk figures are estimated from parameter count — review and refine.
Qwen-Image-Edit-2509
Auto-added from Hugging Face (Qwen/Qwen-Image-Edit-2509). VRAM/quant/disk figures are estimated from parameter count — review and refine.
Qwen-Image-Edit-2511
Auto-added from Hugging Face (Qwen/Qwen-Image-Edit-2511). VRAM/quant/disk figures are estimated from parameter count — review and refine.
HunyuanImage-3.0-Instruct
Auto-added from Hugging Face (tencent/HunyuanImage-3.0-Instruct). VRAM/quant/disk figures are estimated from parameter count — review and refine.
Wan2.2-I2V-A14B
Auto-added from Hugging Face (Wan-AI/Wan2.2-I2V-A14B). VRAM/quant/disk figures are estimated from parameter count — review and refine.
ideogram-4-fp8
Auto-added from Hugging Face (ideogram-ai/ideogram-4-fp8). VRAM/quant/disk figures are estimated from parameter count — review and refine.
Wan2.2-T2V-A14B
Auto-added from Hugging Face (Wan-AI/Wan2.2-T2V-A14B). VRAM/quant/disk figures are estimated from parameter count — review and refine.
Krea-2-Turbo
Auto-added from Hugging Face (krea/Krea-2-Turbo). VRAM/quant/disk figures are estimated from parameter count — review and refine.
FLUX.2-klein-base-9B
Auto-added from Hugging Face (black-forest-labs/FLUX.2-klein-base-9B). VRAM/quant/disk figures are estimated from parameter count — review and refine.
FastVideo-FastH3-4-step-Preview-v1-VSA-DataFree
Auto-added from Hugging Face (FastVideo/FastVideo-FastH3-4-step-Preview-v1-VSA-DataFree). VRAM/quant/disk figures are estimated from parameter count — review and refine.
krea-realtime-video
Auto-added from Hugging Face (krea/krea-realtime-video). VRAM/quant/disk figures are estimated from parameter count — review and refine.
FLUX.2-klein-9B-GGUF
Auto-added from Hugging Face (unsloth/FLUX.2-klein-9B-GGUF). VRAM/quant/disk figures are estimated from parameter count — review and refine.
Krea-2-Raw
Auto-added from Hugging Face (krea/Krea-2-Raw). VRAM/quant/disk figures are estimated from parameter count — review and refine.
FLUX.2-klein-9b-fp8
Auto-added from Hugging Face (black-forest-labs/FLUX.2-klein-9b-fp8). VRAM/quant/disk figures are estimated from parameter count — review and refine.
FLUX.2-klein-9b-kv
Auto-added from Hugging Face (black-forest-labs/FLUX.2-klein-9b-kv). VRAM/quant/disk figures are estimated from parameter count — review and refine.
MisoTTS
Auto-added from Hugging Face (MisoLabs/MisoTTS). VRAM/quant/disk figures are estimated from parameter count — review and refine.
flux2-klein-9b-uncensored-text-encoder
Auto-added from Hugging Face (ponpoke/flux2-klein-9b-uncensored-text-encoder). VRAM/quant/disk figures are estimated from parameter count — review and refine.
FLUX.2-klein-base-9b-fp8
Auto-added from Hugging Face (black-forest-labs/FLUX.2-klein-base-9b-fp8). VRAM/quant/disk figures are estimated from parameter count — review and refine.
JoyAI-Echo
Auto-added from Hugging Face (jdopensource/JoyAI-Echo). VRAM/quant/disk figures are estimated from parameter count — review and refine.
Wan2.2-T2V-A14B-Diffusers
Auto-added from Hugging Face (Wan-AI/Wan2.2-T2V-A14B-Diffusers). VRAM/quant/disk figures are estimated from parameter count — review and refine.
VibeVoice-Large-Q8
Auto-added from Hugging Face (FabioSarracino/VibeVoice-Large-Q8). VRAM/quant/disk figures are estimated from parameter count — review and refine.
FastVideo-FastH3-8-Step-V2
Auto-added from Hugging Face (FastVideo/FastVideo-FastH3-8-Step-V2). VRAM/quant/disk figures are estimated from parameter count — review and refine.
acestep-captioner
Auto-added from Hugging Face (ACE-Step/acestep-captioner). VRAM/quant/disk figures are estimated from parameter count — review and refine.
LTX-2.5-Diffusers
Auto-added from Hugging Face (Lightricks/LTX-2.5-Diffusers). VRAM/quant/disk figures are estimated from parameter count — review and refine.
Eyes_Direction_Lora_Flux2Klein9B
Auto-added from Hugging Face (eric-venti-seeds/Eyes_Direction_Lora_Flux2Klein9B). VRAM/quant/disk figures are estimated from parameter count — review and refine.
MOSS-SoundEffect
Auto-added from Hugging Face (OpenMOSS-Team/MOSS-SoundEffect). VRAM/quant/disk figures are estimated from parameter count — review and refine.
Qwen-Image-2.1-PE-T2I
Auto-added from Hugging Face (Qwen/Qwen-Image-2.1-PE-T2I). VRAM/quant/disk figures are estimated from parameter count — review and refine.
LTX-2.3-22B-DISTILLED-1.1-GGUF
Auto-added from Hugging Face (Abiray/LTX-2.3-22B-DISTILLED-1.1-GGUF). VRAM/quant/disk figures are estimated from parameter count — review and refine.
Flux2-Klein-9B-True-V3
Auto-added from Hugging Face (wikeeyang/Flux2-Klein-9B-True-V3). VRAM/quant/disk figures are estimated from parameter count — review and refine.
pocket-tts-farsi-v2
Auto-added from Hugging Face (mehdi-hf/pocket-tts-farsi-v2). VRAM/quant/disk figures are estimated from parameter count — review and refine.
Kokoro-7M-Distill
Auto-added from Hugging Face (oddadmix/Kokoro-7M-Distill). VRAM/quant/disk figures are estimated from parameter count — review and refine.
JoyAI-Image-Edit-Plus-Diffusers
Auto-added from Hugging Face (jdopensource/JoyAI-Image-Edit-Plus-Diffusers). VRAM/quant/disk figures are estimated from parameter count — review and refine.
LTX-2.3
Image-to-video model with synchronized audio generation. The official card describes LTX-2.3 as a significant update to LTX-2 with improved audio and visual quality and prompt adherence.
Sulphur 2
Uncensored 9B text-to-video and image-to-video model built on LTX 2.3 with integrated prompt enhancer and distill LoRAs for faster sampling on local hardware.[3][4][9][14][15]
LTX-2.5
Multimodal generative model that creates synchronized video and audio from text, image, and video inputs for local execution and fine-tuning.[1][5][12]
Supertonic 3
Multilingual ONNX text-to-speech model running locally, supporting 31 languages with expression tags like laugh and breath.
AuK
AuK is a 1.5B open-source speech foundation model for text-to-speech and speech editing via natural-language instructions.
SCAIL-2
Open-source model for end-to-end controlled character animation. The official docs say it supports character animation and in-video character replacement, with 512p and 704p workflows.
Inflect-Nano-v1
Tiny English text-to-speech model providing a full text-to-waveform pipeline under 5M parameters at 24 kHz, with a single male voice.
LTX-2.3 FP8
DiT-based generative model that produces synchronized audio-video clips from text or image inputs, released as FP8 checkpoints for local inference.
SANA-WM
A 2.6B-parameter image-to-video world model that generates 720p, minute-scale videos with 6-DoF camera control. This repo ships the bidirectional checkpoint and refiner components.
Supra2-IMG
Tiny 100M-parameter diffusion transformer text-to-image model trained from scratch on synthetic data, using Flan-T5-Base text encoder and SD-VAE-FT-MSE VAE at 256×256 resolution.
LTX-2.5-Pre-Trained
Open-weights 22B world model that turns images plus text into short videos with bundled audio and video VAEs.
ZeroTTS
Vietnamese zero-shot text-to-speech model in ONNX format, optimized for real-time local CPU inference with bundled voices and a MOSS codec decoder.
Stable Audio 3 Optimized
Optimized Stable Audio 3 checkpoints for music and sound effects generation, including small and medium DiT transformers plus SAME-L autoencoder variants for faster local inference.
CausalWM
CausalWM is Aether AI’s 16B-parameter causal world model that generates realistic future video frames from input images by explicitly modeling object motion and 3D geometry changes.
Krea 2
Krea 2 is a high-end text-to-image diffusion model with Raw and Turbo variants, using a Qwen3-VL-4B text encoder and Krea Image VAE for ComfyUI workflows.
MiniMax-H3
Auto-added from Hugging Face (MiniMaxAI/MiniMax-H3). VRAM/quant/disk figures are estimated from parameter count — review and refine.
Bernini-R
Standalone DiT-based video renderer from ByteDance for text-to-video and related video generation tasks, derived from Wan2.2-T2V-A14B and released with open weights under Apache-2.0.
LongCat-Video
13.58B-parameter open-source diffusion transformer video generator supporting text-to-video, image-to-video and video continuation at up to 720p 30 fps.
No magic. Just math you can audit.
We compute every model’s VRAM at FP16, FP8, and Q4 GGUF; add a 10% KV-cache safety margin and a 0.5 GB runtime overhead; then check fit against your GPU. Speed comes from a bandwidth-bound model (steps/s ≈ bandwidth ÷ model size × efficiency). All client-side.
See the formulas