# AI models for 3D pipelines

> A directory of 214 AI models from 54 makers across 3D generation, 3D processing, image, language, and audio.

Hydrilla runs one of these models: BlueFox, Hydrilla's 3D model (current version BlueFox 1). https://hydrilla.ai/bluefox
Every other entry is a reference listing. No partnership, endorsement, or availability in Hydrilla is implied. Video models are out of scope.

## 3D generation models (26)

3D generation models turn a text prompt or an image into a mesh. They differ most in what arrives with the geometry: UVs, PBR texture maps, separate parts, and clean topology.

### Hydrilla AI (runs in Hydrilla)

BlueFox, Hydrilla's 3D model, returns a mesh split into parts, UV-unwrapped, with PBR maps, from a typed prompt or an uploaded image.

- BlueFox 1: Text or image to 3D

[About BlueFox](https://hydrilla.ai/bluefox)

### Tripo

Text, image, and sketch to 3D, with low-poly and part-based variants.

- Tripo 3.5: Text or image to 3D
- Tripo 3.0: Text or image to 3D
- Tripo3D v3.1: Text or image to 3D
- Tripo P2: Text or image to 3D
- Tripo P1: Text or image to 3D
- Tripo Smart Low Poly: Low-poly generation
- Tripo Generate Parts: Part-based generation

### Meshy

Text and image to 3D, with a lighter tier and a topology-aware variant.

- Meshy v7.1: Text or image to 3D
- Meshy v6: Text or image to 3D
- Meshy 6 Lite: Text or image to 3D
- Meshy-T2.5 Smart Topology: Topology-aware generation
- Meshy Figure Prototype: Figure generation

### Tencent

The Hunyuan3D family, plus Pixal3D from Tencent ARC, which lifts pixel features into 3D for pixel-aligned geometry and PBR textures.

- Hunyuan3D 3.1: Text or image to 3D
- Hunyuan3D 3.0 LowPoly: Low-poly generation
- Hunyuan3D Rapid: Text or image to 3D
- Pixal3D: Image to 3D

### Microsoft

Open-weights checkpoints from Microsoft Research that teams host on their own GPUs.

- TRELLIS 2: Image to 3D

### Hyper3D

Rodin, the 3D model from Hyper3D by Deemos.

- Hyper3D Rodin 2.5: Text or image to 3D

### Hi3D

High-resolution image to 3D in two quality tiers.

- Hi3D v3.0 Quality: Image to 3D
- Hi3D v3.0 Master: Image to 3D

### Meta

3D reconstruction of objects from a single image.

- SAM 3D: Image to 3D reconstruction

### DreamTech

- Direct3D-S2: Image to 3D

### GAP Lab, CUHK-Shenzhen

A research model for accurate multi-view object reconstruction through generation.

- ReconViaGen 0.5: Multi-view reconstruction

### Atlas

- Titan v1: Text or image to 3D
- 404—GEN: Text or image to 3D

## 3D processing, texture, and rigging models (26)

3D processing models work on a mesh that already exists: they retopologize it, unwrap UVs, paint textures, split it into parts, rig it, or animate it.

### Meshy

- Meshy Remesh: Retopology
- Meshy UV Unwrap: UV unwrap
- Meshy Texture v7: Texturing
- Meshy Texture v6: Texturing
- Meshy N1 Texture: Texturing
- Meshy Rigging: Rigging
- Meshy Text to Motion: Animation
- Meshy Auto Split: Part splitting

### Tripo

- Tripo Texture 3.5: Texturing
- Tripo Texture 3.0: Texturing
- Tripo Retopology: Retopology
- Tripo Smart UV: UV unwrap
- Tripo Segmented Parts: Part splitting
- Tripo Mesh Segmentation: Segmentation
- Tripo Rigging v1.0 / v3.0: Rigging

### Tencent

- Hunyuan UV Unfold: UV unwrap
- Hunyuan Smart Topology: Retopology
- Hunyuan3D Parts: Part splitting
- Tencent 3D Texture Edit: Texture editing

### Hyper3D

- Hyper3D Bang: Part splitting

### Hi3D

- Hi3D Split: Part splitting
- Hi3D v3.0 Texture: Texturing

### fal

Seamless, tiling PBR material sets from text or a starting image.

- PATINA PBR Materials: PBR materials

### NVIDIA

A kinematic motion diffusion model controlled by text and keyframe constraints.

- Kimodo Text to Motion: Animation

### Atlas

- Atlas Remesh: Retopology
- Atlas 1 Rigging: Rigging

## Image models (94)

Image models generate and edit pictures from a prompt. In a 3D pipeline they produce the concept art and the reference image that image-to-3D starts from, and they upscale, cut out, and vectorize source art.

### Google DeepMind

The Nano Banana line, also listed under its Gemini image names.

- Nano Banana 2.1: Generate, edit
- Nano Banana 2: Generate, edit
- Nano Banana 2 Lite: Generate, edit
- Nano Banana Pro: Generate, edit
- Gemini 3 Pro Image: Generate, edit
- Gemini 3.1 Flash Image: Generate, edit
- Gemini 3.1 Flash Lite Image: Generate, edit
- Gemini 3 Pro Upscale: Upscale
- Gemini 3.1 Flash Upscale: Upscale

### OpenAI

GPT Image, with a fast Flare tier and a precision Sunburst tier.

- GPT Image 2.6 Flare: Generate, edit
- GPT Image 2.6 Sunburst: Generate, edit
- GPT Image 2.5 Flare: Generate, edit
- GPT Image 2.5 Sunburst: Generate, edit
- GPT Image 2: Generate, edit

### Black Forest Labs

The FLUX family, from hosted flagship tiers to small open-weights models.

- FLUX 3 Image: Generate, edit
- FLUX 3 Image Klein: Generate, edit
- FLUX.2 Max: Generate, edit
- FLUX.2 Pro: Generate, edit
- FLUX.2 Flex: Generate, edit
- FLUX.2 Klein 4B: Generate
- FLUX.1 Dev: Generate
- FLUX.1 Schnell: Generate

### Alibaba

Qwen Image for generation and editing, with strong text rendering, plus the fast Z-Image Turbo.

- Qwen Image 3: Generate, edit
- Qwen Image 3 Pro: Generate, edit
- Qwen Image 2.1 Pro: Generate, edit
- Qwen Image 2.1 Turbo: Generate, edit
- Qwen Image 2 Pro: Generate, edit
- Qwen-Image: Generate, open weights
- Qwen Image Layered: Layered output
- Qwen Camera Control: Camera edit
- Z-Image Turbo: Generate

### ByteDance

Seedream for generation and editing, plus upscaling, inpainting, and depth estimation models.

- Seedream 5.5: Generate, edit
- Seedream 5.0 Pro: Generate, edit
- Seedream 5.0 Lite: Generate, edit
- Seedream 5.0 Flash: Generate, edit
- Seedream 4.5: Generate, edit
- SeedVR2 Upscaler: Upscale
- OneReward: Inpainting
- Depth Anything V3: Depth estimation
- Depth Anything V2: Depth estimation

### Recraft

Design-first models with brand colours as an input and real SVG output in the vector variants.

- Recraft V4.1: Generate
- Recraft V4.1 Pro: Generate
- Recraft V4.1 Flash: Generate
- Recraft V4.1 Utility: Generate
- Recraft V4.1 Utility Pro: Generate
- Recraft V4: Generate
- Recraft V4 Pro: Generate
- Recraft V4 Vector: Vector
- Recraft V3: Generate, edit
- Recraft Text to Vector: Vector
- Recraft Vectorize: Vectorize
- Recraft Upscale: Upscale

### Ideogram

Built for design and typography, with region editing and SVG output.

- Ideogram 4.5: Generate, edit
- Ideogram 4.5 Precise Edit: Region edit
- Ideogram 4.0: Generate
- Ideogram Vector: Vector
- Ideogram Vectorize: Vectorize
- Ideogram Remove Background: Background removal

### Krea

- Krea 3: Generate, edit
- Krea 2 Large: Generate
- Krea 2 Medium: Generate
- Krea 2 Medium Turbo: Generate
- Krea 2 Turbo: Generate

### xAI

- Grok Imagine 2.0: Generate, edit
- Grok Imagine: Generate

### Bria

Image models trained on licensed data.

- Bria FIBO 2.0: Generate, edit
- Bria FIBO 1.5: Generate
- Bria RMBG 2.0: Background removal
- Bria Guided Remove Background: Background removal

### Microsoft

- MAI-Image-2.5: Generate
- MAI-Image-2.5 Pro: Generate, edit

### Kuaishou

The image models from the Kling line.

- Kling Image O3: Generate, edit
- Kling Image v3: Generate, edit

### MiniMax

- MiniMax H3 Image: Generate, edit
- MiniMax H3 Image Highspeed: Generate, edit

### Tencent

- Hunyuan Image 3.0 Instruct: Generate, edit

### Baidu

- ERNIE Image: Generate
- ERNIE Turbo: Generate

### Meta

- Muse Image: Generate, edit

### Luma AI

- Luma Uni-1: Generate, edit

### HiDream.ai

Open-weight image models.

- HiDream-I1 Full: Generate
- HiDream-O1: Generate, edit

### Topaz Labs

- Topaz Precision Image Upscale: Upscale
- Topaz Image Upscale: Upscale

### Quiver

- Quiver Arrow 1.1: Vector
- Quiver Arrow 1.1 Max: Vector

### OpenMOSS

An open model for text and layout that plans the layout before it draws.

- MOSS-Image: Generate

### ImagineArt

- ImagineArt 1.5 Pro: Generate

### Pruna AI

- P-Image: Generate

### Hi3D

- Hitem3D Relief: Relief from an image

### Pixelcut

- Pixelcut Background Remover: Background removal

### Smoretalk

- Rembg Enhance: Background removal

### Feyn

- FeyNoBG: Background removal

### Visioncortex

- Image2SVG: Vectorize

## Language models (28)

Language models read, reason, and write text and code. In a 3D pipeline they write and expand prompts, plan asset lists, and run the AI agents that call generation tools.

### OpenAI

- GPT-6 Astra: Language
- GPT-6 Sol: Language
- GPT-6 Luna: Language
- GPT-6.1 Sol: Language
- GPT-6.1 Terra: Language
- GPT-OSS 120B: Language, open weights
- GPT-OSS 20B: Language, open weights

### Google DeepMind

- Gemini 4 Argon: Language
- Gemini 3.8 Flash: Language
- Gemini 3.5 Flash Lite: Language
- Gemini 3.1 Pro Preview: Language
- Gemini 3.1 Flash: Language
- Gemini 3.1 Flash Lite: Language
- Gemini 3 Pro: Language

### Anthropic

- Claude Fable 5.2: Language
- Claude Opus 5.5: Language
- Claude Sonnet 5.5: Language
- Claude Haiku 5.5: Language
- Claude Fable 5.1: Language
- Claude Haiku 4.5: Language

### xAI

- Grok 4.7: Language
- Grok Build 0.1: Language, coding

### ByteDance

- Seed 2.1 Turbo: Language
- Seed 2.0 Lite: Language
- Seed 2.0 Mini: Language

### DeepSeek

- DeepSeek V4 Pro: Language
- DeepSeek V4.1 Flash: Language

### Z.ai

- GLM 5.3 Flash: Language

## Audio and voice models (40)

Audio models generate speech, music, and sound effects, and transcribe recordings. They give voices to characters and sound to the scenes a 3D team builds.

### ElevenLabs

- Eleven v4: Text to speech
- Eleven v4 Turbo: Text to speech
- Eleven v3: Text to speech
- Eleven Flash v2.5: Text to speech
- ElevenLabs Music v2: Music
- ElevenLabs Music v2.5 Extend: Music
- ElevenLabs SFX v2: Sound effects
- ElevenLabs Scribe v2: Transcription
- ElevenLabs Voice Changer: Voice conversion

### Google DeepMind

- Lyria 3.5: Music
- Lyria 3 Clip: Music
- Gemini 3.8 Flash TTS: Text to speech
- Gemini 3.8 Flash-Lite TTS: Text to speech
- Gemini 3.1 Flash TTS: Text to speech
- Gemini 3.5 Transcribe: Transcription

### MiniMax

- MiniMax Speech 2.8 HD: Text to speech
- MiniMax Speech 02: Text to speech
- MiniMax Music 3: Music
- MiniMax Music Cover: Music

### ByteDance

- Seed TTS 2.0: Text to speech
- Seed Audio 1.0: Audio generation
- Seed ASR 2.0 Flash: Transcription

### HeyGen

- HeyGen Voice: Text to speech
- HeyGen Voice Turbo: Text to speech

### Sonilo

- Sonilo Music: Music
- Sonilo Sound Effects: Sound effects

### Stability AI

- Stable Audio 3: Music and audio

### xAI

- Grok TTS: Text to speech

### Alibaba

- Qwen Audio 3 TTS: Text to speech

### Microsoft

- MAI-Voice 2.1: Text to speech

### Mistral AI

- Voxtral Mini TTS 3.0: Text to speech

### Fish Audio

- Fish Audio Drama 3: Text to speech

### Inworld

- Inworld Realtime TTS 2.0: Text to speech

### AssemblyAI

- AssemblyAI TTS 1: Text to speech

### hexgrad

- Kokoro: Text to speech, open weights

### Bilibili

- IndexTTS 2.0: Text to speech

### ACE Studio and StepFun

- ACE-Step: Music

### Mirelo

- Mirelo SFX 2.0: Sound effects

### VEED

- VEED Clean Audio: Audio cleanup

### Kotoba Technologies

- Kotoba STT: Transcription

## Questions

### Which AI model does Hydrilla use to generate 3D models?

Hydrilla runs BlueFox, Hydrilla's 3D model, built by Hydrilla AI. The current version is BlueFox 1. It turns a typed prompt or an uploaded image into a mesh split into parts, UV-unwrapped, with PBR maps.

### Which AI models can generate a 3D model from text or an image?

3D generation models include BlueFox from Hydrilla AI, Tripo, Meshy, Hunyuan3D and Pixal3D from Tencent, TRELLIS from Microsoft, Rodin from Hyper3D, Hi3D, SAM 3D from Meta, and Direct3D-S2. They differ in whether the mesh arrives with UVs, PBR maps, separate parts, and clean topology.

### Are the other models on this page available in Hydrilla?

No. BlueFox is the model that runs in Hydrilla. Every other model here is a reference listing with no partnership, endorsement, or availability implied. Confirm details on each maker's own site.

### Which image model should I use to make a reference image for image to 3D?

Any image model that gives you one clear subject on a plain background, evenly lit, will do. Models with reliable editing help when you need to clean up a reference before you spend on 3D. Hydrilla's text-to-3D flow also makes a reference image for you to approve first.

### Can a language model generate 3D assets in Hydrilla?

Yes, through an agent. Hydrilla's MCP server lets agents such as Claude Code, Cursor, and Codex generate models and send them into an open Blender, Maya, 3ds Max, Unity, Unreal, or Godot scene. Inside Hydrilla, Hydrilla Agent is the AI 3D agent that builds and changes your Blender scene in chat.

### Why does this directory leave out video models?

It covers the model types that feed a 3D asset pipeline: 3D generation and processing, image, language, and audio. Video generation is outside that scope.

### What is the difference between a 3D generation model and a 3D processing model?

A 3D generation model creates a new mesh from a text prompt or an image. A 3D processing model works on a mesh that already exists: it retopologizes, unwraps UVs, textures, splits the mesh into parts, rigs, or animates it.

### Which AI models can texture, retopologize, or rig a 3D model?

Texturing models include Meshy Texture, Tripo Texture, and Hi3D Texture. Retopology models include Meshy Remesh, Tripo Retopology, and Hunyuan Smart Topology. Rigging models include Meshy Rigging and Tripo Rigging. PATINA from fal generates PBR materials, and Kimodo from NVIDIA generates motion.

### Which AI image models are available today?

Image model families include Nano Banana from Google DeepMind, GPT Image from OpenAI, FLUX from Black Forest Labs, Qwen Image from Alibaba, Seedream from ByteDance, Recraft, Ideogram, Krea, Grok Imagine from xAI, and Bria FIBO. Most generate and edit; some specialize in vectors, upscaling, or background removal.

### Which language models do AI agents use?

Agents run on language models such as GPT from OpenAI, Gemini from Google DeepMind, Claude from Anthropic, Grok from xAI, DeepSeek, and GLM from Z.ai. The agent picks the tools; the language model does the reasoning and writes the calls.

### How current is this list?

The listings reflect November 2026. Model names and versions change quickly, so treat each maker's own site as the source of truth.

## How this directory is compiled

Names and makers come from each maker's public pages and from public model catalogs. We give no scores or rankings.

Listings as of November 2026. Model names and logos belong to their owners and are shown for identification only.
