CPU3DAI tools for 3D, video and audio

AI tools with open source under 24 GB of VRAM

VRAM thresholds, backends and licenses come from the tool fact sheets: every value is collected by code from an official source and stamped with the date it was checked. Requirements marked with an asterisk * are the authors’ own description, not our measurement. Details are on each tool’s page.

Showing 20 of 68

ToolTypeRunsVRAMBackendsFormats
Gaussian Splatting (3DGS)3Dopen sourcefrom 24 GB *CUDA
Hunyuan3D 2.1image-to-3d3Dopen sourcefrom 10 GB (10–29 GB) *
MeshAnything3Dopen sourcefrom 7 GB *CUDA
Stable Fast 3Dimage-to-3d3Dopen sourcefrom 6 GB *CUDAApple MPSglb
threestudio3Dopen sourcefrom 6 GB *CUDAobj
TripoSRimage-to-3d3Dopen sourcefrom 6 GB *CUDA
Zero123++3Dopen sourcefrom 5 GB *
ACE-Steptext-to-audioAudioopen sourcefrom 8 GB *
Barktext-to-speechAudioopen sourcefrom 12 GB *CUDA
ChatTTStext-to-audioAudioopen sourcefrom 4 GB *
MMAudioAudioopen sourcefrom 6 GB *CUDA
MOSS-SoundEffect without CUDA (Mac and AMD)Audioopen sourcefrom 8 GB (8–10 GB) *Apple MPSAMD ROCm
Allegrotext-to-videoVideoopen sourcefrom 9.3 GB (9.3–27.5 GB) *CUDA
AnimateDiffVideoopen sourcefrom 13 GB *
FramePackVideoopen sourcefrom 6 GB *CUDA
LTX-Videoimage-to-videoVideoopen sourcefrom 1 GB *Apple MPS
Open-Sora-PlanVideoopen sourcefrom 24 GB *
Pyramid Flowtext-to-videoVideoopen sourcefrom 8 GB *Apple MPS
SkyReelsVideoopen sourcefrom 18.5 GB *CUDA
Wantext-to-videoVideoopen sourcefrom 8.19 GB *

Requirements not stated by the authors

The official sources for these tools have no data for the selected filter — that means “unknown”, not “won’t work”.

3DTopia3Dopen sourcenot statedCUDAglb
CRM3Dopen sourcenot statedobj
DreamGaussian3Dopen sourcenot stated
Era3D3Dopen sourcenot stated
Hunyuan3D 2.1 without CUDA (Mac and AMD)3Dopen sourcenot statedApple MPSAMD ROCmobj, glb, ply, stl, fbx, dae, 3mf
InstantMeshimage-to-3d3Dopen sourcenot statedCUDAobj
LGMtext-to-3d3Dopen sourcenot stated
Michelangelo3Dopen sourcenot stated
MVDream3Dopen sourcenot stated
Nerfstudio3Dopen sourcenot statedCUDA
One-2-3-453Dopen sourcenot stated
OpenLRMimage-to-3d3Dopen sourcenot statedCUDAobj
Point-E3Dopen sourcenot stated
Shap-E3Dopen sourcenot stated
TRELLISimage-to-3d3Dopen sourcenot statedCUDAglb
Unique3D3Dopen sourcenot statedCUDA
Wonder3D3Dopen sourcenot stated
CosyVoicetext-to-speechAudioopen sourcenot stated
DiffRhythmAudioopen sourcenot stated
F5-TTStext-to-speechAudioopen sourcenot statedAMD ROCm
Fish Speechtext-to-speechAudioopen sourcenot statedCUDA
MOSS-SoundEffecttext-to-audioAudioopen sourcenot stated
MusicGentext-to-audioAudioopen sourcenot stated
OpenVoicetext-to-speechAudioopen sourcenot stated
Stable Audio Opentext-to-audioAudioopen sourcenot stated
XTTStext-to-speechAudioopen sourcenot statedCUDA
YuEtext-generationAudioopen sourcenot stated
CogVideoXtext-to-videoVideoopen sourcenot statedCUDA
LatteVideoopen sourcenot statedCUDA
Open-SoraVideoopen sourcenot stated
Stable Video Diffusionimage-to-videoVideoopen sourcenot stated
VideoCrafterVideoopen sourcenot stated

Fact sheets are re-checked monthly; every value on a tool page has a source link and a check date. The asterisk * means “as described by the author”: a developer’s statement, not an independent measurement.