Release · Models · 3D and audio
AI 3D model generator API: Tripo, Meshy and Hunyuan 3D, plus music and sound effects
FlyMy.AI now runs 14 AI 3D model generators from Tripo, Meshy and Tencent's Hunyuan 3D: a photo, several views or a prompt becomes a textured 3D model, a character gets a rig and a mesh gets new textures. Four music and sound models came with them. One API key and one MCP reach all of them, from Claude Code, Cursor, an agent or your own code. What each model does, our test runs with their prompts and prices, and how to call them.
What is an AI 3D model generator?
An AI 3D model generator turns a picture or a description into a 3D model: a mesh of triangles or quads that describes the shape, and textures that paint it. The model comes back as a file that 3D tools open, usually GLB (binary glTF) for the web, game engines and Blender, or FBX for animation tools. Three kinds of input are common:
- Image to 3D: one photo of a single object. The model builds what it sees and guesses the sides it cannot.
- Multi-view to 3D: two to eight photos of the same object from different angles, so the back and the sides come from real views.
- Text to 3D: a description of one object. Tripo first draws a picture from the prompt and builds the model from it; its
image_seedinput repeats that picture.
Two more jobs finish a model. Rigging adds a skeleton so a character can walk, run or wave in an engine. Retexturing paints new materials on a mesh you already have. Two settings decide where a model can go: PBR materials (base color plus metallic, roughness and normal maps) make it react to light like the real thing, and the polygon count decides what can render it. A mesh of a million triangles suits a product render; a game on a phone wants a few thousand.
The 14 AI 3D model generators on FlyMy.AI
Since 18 November 2026 these models run on FlyMy.AI. Each one is an endpoint with its own page under every model on FlyMy.AI, where you find every input, the price and requests to paste.
| Model | Input | What you get | Price with defaults |
|---|---|---|---|
| Tripo H3.1 Image to 3D | One photo (image) | A textured GLB with PBR materials and a preview render; FBX with quad topology | $0.30 |
| Tripo H3.1 Multiview to 3D | 2 to 4 photos (images) | The same, with the hidden sides taken from real views | $0.30 |
| Tripo H3.1 Text to 3D | A prompt | A textured GLB with PBR materials | $0.20 |
| Tripo P2 Image to 3D | One photo | A low-poly model of 48 to 50,000 faces, triangles or native quads | $1.10 |
| Tripo P2 Text to 3D | A prompt | The same, from text | $1.10 |
| Meshy 7.1 Image to 3D | One photo | A detailed GLB and FBX; a rigged, animated humanoid on request | $1.20 |
| Meshy 7.1 Multi-Image to 3D | 1 to 4 photos (images) | One model of the object in the photos | $1.20 |
| Meshy 7.1 Text to 3D | A prompt | A textured GLB and FBX; a cheaper untextured preview mode | $1.20 |
| Meshy T2.5 Image to 3D | One photo | A low-poly triangle mesh with separated parts, up to 30,000 triangles | $0.60 |
| Meshy T2.5 Text to 3D | A prompt | The same, from text | $0.60 |
| Hunyuan 3D 3.1 Pro | A front photo and up to 7 more views | A detailed GLB of 40,000 to 1,500,000 faces | $0.375 |
| Hunyuan 3D 3.1 Rapid | One photo | A quick OBJ with its material and texture files, for drafts and batches | $0.225 |
| Meshy Rigging | A GLB character (model) | The rigged GLB and FBX with walking and running clips | $0.20 |
| Meshy 7 Retexture | A 3D model and a style prompt or picture | The same mesh with new textures, GLB and FBX | $0.30 |
Prices are per model at the default settings, from FlyMy.AI billing; What it costs lists what changes them.
Test 1: one photo to a textured 3D model
We made the photo with Seedream 5.5 on FlyMy.AI, one 2048x2048 image for $0.035, with this prompt:
A charming handcrafted ceramic fox figurine with glossy copper-orange glaze, cream chest and tail tip, standing alert, full body, centered, three-quarter view, soft studio lighting, pure plain white background, no shadow, no text
Then we sent the photo, unchanged, to Tripo H3.1 Image to 3D with the default settings: standard textures with PBR materials, standard geometry, triangles. The price for these settings is $0.30, and the model answered in 112 seconds with two files:
model: a 44.6 MB GLB with one mesh of 750,254 vertices and 1,472,140 triangles, and three texture maps: base color, metallic and roughness, and a normal map.preview: a 512x512 render of the model, the picture on the right below.
The textures carry the copper-orange glaze, the cream chest and tail tip and the dark eyes and nose of the photo;
the glaze on the model is paler and patchier than in the photo. Tripo also built the sides the photo does not
show. The GLB is large: at 44.6 MB it suits a render or an editor more than a web page, and
face_limit asks for fewer faces when you need a lighter file.
This page has no 3D viewer and the GLB is too large to host here. To get the file, run the same request: the
answer's model output is a link to the GLB. The links in an answer are temporary, so download the
files you keep.
Turn this photo into a textured 3D model with Tripo H3.1 and send me the GLB file: https://example.com/ceramic-fox.png
Test 2: one prompt, two 3D models
The same three words, a small wooden chair, went to two text to 3D models, both without textures so the shapes compare:
- Tripo H3.1 Text to 3D with
textureandpbroff: $0.10, 203 seconds, a 41.7 MB GLB of 753,862 vertices and 1,465,686 triangles. - Meshy T2.5 Text to 3D with
modepreview, which stops at the untextured mesh: $0.20, 80 seconds, a 0.28 MB GLB of 7,966 vertices and 15,700 triangles, and an FBX copy of 0.37 MB.
Tripo H3.1 carved the wood grain into the geometry; Meshy T2.5 made a clean low-poly chair with 93 times fewer triangles in a file 146 times smaller, the kind of mesh a mobile game or a web scene can load as it is. Pick by where the model goes, not by which looks richer in a render. Download Meshy T2.5's chair as GLB (0.3 MB).
Which 3D model to use
- A product or a prop from one photo: Tripo H3.1 Image to 3D ($0.30). For many quick drafts, Hunyuan 3D 3.1 Rapid ($0.225).
- The back and the sides matter: send more views to Tripo H3.1 Multiview to 3D (up to four), Meshy 7.1 Multi-Image to 3D (up to four) or Hunyuan 3D 3.1 Pro (the front and up to seven more).
- A game asset with a polygon budget: Tripo P2 (48 to 50,000 faces, native quads) or Meshy T2.5 (up to 30,000 triangles, parts already separated).
- A character that moves: Meshy 7.1 with
enable_riggingandenable_animation, or Meshy Rigging for a GLB character from any model. - A new look for a model you have: Meshy 7 Retexture, from a style prompt or a picture, up to 8K.
- From a description: Tripo H3.1 Text to 3D ($0.20, or $0.10 without textures), Meshy 7.1 Text to 3D with a cheaper preview mode, or the low-poly Tripo P2 and Meshy T2.5.
Independent votes rank the same models high. On the Top3D AI leaderboard, Tripo H3.1 placed first for textured models and for geometry in the October 2026 season, and in the November season, still running when we checked on 17 November 2026, Tripo H3.1, Meshy 7.1 and Hunyuan 3D 3.1 were first, second and fourth overall.
How to call the 3D model API
Four ways in, from a sentence in your assistant to a request from your own code.
1. From Claude Code, Claude, ChatGPT or Cursor
Add the FlyMy.AI MCP at https://mcp-agents.flymy.ai/mcp and sign in:
- Claude Code: run
claude mcp add --transport http flymyai https://mcp-agents.flymy.ai/mcp, then/mcpto sign in. - Claude and ChatGPT: Settings, Connectors, Add custom connector, then paste the URL and sign in.
- Cursor: add the same URL as a remote MCP server.
Then ask in plain words, with a public link to your file:
Turn this photo of my ceramic fox into a textured 3D model with Tripo H3.1 and send me the GLB: https://example.com/ceramic-fox.png
Make a low-poly wooden barrel for a mobile game with Meshy T2.5, about 5,000 triangles, and send me the GLB.
Rig this character for Unreal Engine 5 with Meshy Rigging and send me the walking clip: https://example.com/character.glb
Your assistant reads the model's inputs with get_model_schema and calls run_model with:
{
"endpoint_id": "flymyai/tripo-h3_1-image-to-3d",
"input": {
"image": "https://example.com/ceramic-fox.png"
}
}
A character to rig goes in model, with the skeleton your engine expects:
{
"endpoint_id": "flymyai/meshy-rigging",
"input": {
"model": "https://example.com/character.glb",
"skeleton_template": "ue5_mannequin"
}
}
A 3D run takes minutes; if it is still going, your assistant reads the result with get_model_result.
2. From your code: the REST API
Send form fields, not JSON, with your FlyMy.AI API key in the X-API-KEY header. Your keys are on the
Developers page; keep yours in an environment variable, never in
code you share. Because a 3D model takes minutes, queue the run on the async route:
curl -fsS -X POST https://api.flymy.ai/api/v1/flymyai/tripo-h3_1-image-to-3d/predict/async/ \
-H "X-API-KEY: $FLYMYAI_API_KEY" \
-F "image=@ceramic-fox.png"
The answer is a prediction_id. Read the result with it:
curl -fsS "https://api.flymy.ai/api/v1/flymyai/tripo-h3_1-image-to-3d/predict/async/result/?request_id=<prediction_id>" \
-H "X-API-KEY: $FLYMYAI_API_KEY"
The result lists the model's outputs, each file as a list of links with its type and name. For Tripo H3.1 the outputs read:
{
"model": [
{
"url": "https://...",
"content_type": "model/gltf-binary",
"file_name": "model.glb"
}
],
"preview": [
{
"url": "https://...",
"content_type": "image/webp",
"file_name": "preview.webp"
}
],
"task_id": "...",
"calculated_price": "0.300000"
}
A link to a file you already host works in place of an upload: -F 'image=https://example.com/ceramic-fox.png'.
Inputs that take several files, such as images for the multi-view models, take a JSON list of links.
Every model's page lists its inputs, and Model endpoints
in the docs covers the routes.
3. Inside an agent
A FlyMy.AI agent runs the same models as part of its work: a store agent can turn each new product photo into a GLB for the product page, and a game studio's agent can rig each new character before it goes to the engine. Name the model in the instruction, or describe the job and let the agent choose.
4. In the playground
Open Tripo H3.1 Image to 3D in the app, upload a photo and run it; the result has the preview and the 3D file to download.
Songs, instrumentals and sound effects through the API
Four audio models came with the 3D release. They take a prompt, and for a song your lyrics, and answer with an audio file:
- Mureka V9.5 Song writes and sings a complete song: send
lyrics(with section tags such as [Verse] and [Chorus]) and a style note inprompt, or only a prompt and it writes the words too.genderpicks the voice, and for a song from a promptstylestakes up to four genres. - Mureka V9.5 Instrumental composes background music without vocals from a prompt; the model picks the length.
- Lyria 3.5, Google DeepMind's music model, composes a song with sung lyrics or an instrumental from a description, in eight languages.
- Mirelo SFX 2.0 makes sound effects and ambience from a description, 1 second to 10 minutes, up to four variations at once.
Our runs, with what we sent and what came back:
Mureka V9.5 Song: our lyrics, sung, $0.15
We sent two short sections of lyrics, the style note bright indie pop, warm acoustic guitar, upbeat and gender female:
[Verse] Morning light on the window sill Coffee steaming, the city is still [Chorus] We are just getting started Hearts wide open, never departed
Mureka V9.5 Instrumental: background music, $0.15
The prompt was calm lo-fi piano, slow tempo. The model answered in 62 seconds with a track of 2 minutes 52 seconds; the player has its first 45 seconds.
Lyria 3.5: a jingle that became a song, $0.10
The prompt was short upbeat acoustic guitar jingle. In 46 seconds Lyria 3.5 returned a 62-second song with vocals, and wrote the words itself: "Bright lights, a brand new day begins / With every note, a little magic spins". Say instrumental in the prompt when you want no voice, and ask for a length such as a 30-second track.
Mirelo SFX 2.0: a creaking door, $0.015
The prompt was a wooden door creaking open with duration 1: one second of sound, a WAV file, in
8 seconds, for $0.015. The default 10 seconds cost $0.15, and num_samples asks for up to four takes.
Make three takes of rain on a metal roof, 20 seconds each, with Mirelo SFX 2.0, with no music or voices.
Write a 2-minute indie pop song about a summer road trip with Mureka V9.5, female vocals, and send me the MP3.
Also new: video and image models
The same release added six video endpoints and two image endpoints, and six Kling 4.0 video endpoints followed later the same day:
| Model | What it makes | Price with defaults |
|---|---|---|
| Veo 3.1 Lite (image to video) | Google's low-cost Veo 3.1 tier: 4, 6 or 8 second clips with sound, 720p or 1080p | $0.40 for 8 s at 720p |
| HeyGen Video 1 (image to video) | 5 to 30 second clips with dialogue, ambience and effects, up to 2K | $0.10 for 5 s at 768p |
| FLUX 3 Max (image to video) | A post-trained FLUX 3 Video: 5 to 20 seconds with native sound, 720p or 1080p | $0.85 for 5 s at 720p |
| Kling 4.0 (image and reference to video) | Kuaishou's Kling 4.0: 3 to 30 seconds with stereo sound, at 720p or 1080p | $1.175 for 5 s at 1080p |
| Kling 4.0 Flash (image and reference to video) | The fast tier of Kling 4.0: 3 to 20 seconds at 720p, with sound | $0.47 for 5 s |
| Kling Image V4 (Edit) | 2K images from a prompt, or edits of up to ten of your pictures; single images or a matching series | $0.063 per image |
Then came models priced by the length or size of the files you send: talking videos from a photo and a voice (HeyGen Avatar 6, Sync Lipsync 3 Image to Video), lip sync on a video (Sync Lipsync 3), editing a video from a prompt (Kling 4.0 Video Edit), sound effects for a video (Mirelo SFX 2.0 Video to Audio) and upscalers for pictures (Topaz Image Upscale, SeedVR2) and video (Topaz Starlight, SeedVR2 Video Upscale, ByteDance Video Upscaler). Each has its own page under every model on FlyMy.AI.
What it costs
Every run is billed to your FlyMy.AI balance at a price set by its inputs, known before the run starts; an
assistant can quote it with get_pricing. These are the rules from FlyMy.AI billing:
| Model | What changes the price | Price |
|---|---|---|
| Tripo H3.1 Image to 3D, Multiview | No textures / standard textures / HD textures; detailed geometry; quad topology | $0.20 / $0.30 / $0.40; +$0.20; +$0.05 |
| Tripo H3.1 Text to 3D | No textures / standard / HD; detailed geometry; quad topology | $0.10 / $0.20 / $0.30; +$0.20; +$0.05 |
| Tripo P2 (image or text) | No textures / fast or standard / detailed / extreme textures; quads cost nothing extra | $1.00 / $1.10 / $1.20 / $1.30 |
| Meshy 7.1 (all three) | Untextured / textured; rigging; one animation | $0.80 / $1.20; +$0.20; +$0.12 |
| Meshy T2.5 (image or text) | Untextured / textured; rigging; one animation | $0.20 / $0.60; +$0.20; +$0.12 |
| Meshy Rigging | The rig; with one animation | $0.20; $0.32 |
| Meshy 7 Retexture | 2K or 4K textures / 8K textures | $0.30 / $0.45 |
| Hunyuan 3D 3.1 Pro | Base; PBR materials; extra views (once, however many); your own face_count | $0.375; +$0.15; +$0.15; +$0.15 |
| Hunyuan 3D 3.1 Rapid | Base; PBR materials | $0.225; +$0.15 |
| Mureka V9.5 Song | With your lyrics / from a prompt alone | $0.15 / $0.50 |
| Mureka V9.5 Instrumental | Per track | $0.15 |
| Lyria 3.5 | Per song | $0.10 |
| Mirelo SFX 2.0 | Per second of sound, per variation (10 seconds by default) | $0.015 per second |
At these prices the test runs on this page come to $1.05: $0.035 for the fox photo, $0.30 for its 3D model, $0.10 and $0.20 for the two chairs, $0.15 each for the Mureka song and instrumental, $0.10 for the Lyria song and $0.015 for the creaking door.
What these models do not do
- One object, not a scene. Each run makes one model; for a room or a level, generate the objects one by one.
- The shape comes from one kind of input. Image to 3D, Multiview and Multi-Image endpoints build it from pictures, Text to 3D endpoints from a prompt; Meshy's
texture_promptandtexture_image_urlonly steer the look. - Rigging is for humanoid characters. Meshy does not rig animals or vehicles, and it asks for a textured humanoid of at most 380,000 triangles facing +Z. A dense model, like Tripo H3.1's 1.47 million triangles, needs a lower
face_limitbefore it can be rigged. - Files are large and links expire. A detailed GLB can pass 40 MB; download what you keep.
- Audio limits: Mureka V9.5 Instrumental takes no exact length, Lyria 3.5 takes no reference song, and Mirelo SFX 2.0 makes sound from text only, with no music or speech; to give a video its sound use Mirelo SFX 2.0 Video to Audio or Video to Video.
FAQ
What is the best AI 3D model generator?
It depends on the job. On the Top3D AI leaderboard, Tripo H3.1 placed first for textured models and for geometry in the October 2026 season, and in the running November season Tripo H3.1, Meshy 7.1 and Hunyuan 3D 3.1 were first, second and fourth overall when we checked on 17 November 2026. For one photo, start with Tripo H3.1; for a low-poly game asset, Tripo P2 or Meshy T2.5; for a character you will animate, Meshy 7.1 with rigging; for many quick drafts, Hunyuan 3D 3.1 Rapid. All of them run on FlyMy.AI with one API key.
Can AI turn a photo into a 3D model?
Yes. An image to 3D model builds a textured 3D model from one photo and guesses the sides it cannot see; a multi-view model takes several photos of the same object for a more faithful back and sides. In our test, Tripo H3.1 turned a photo of a ceramic fox into a 44.6 MB GLB with PBR textures in about two minutes, for $0.30.
How do I make a 3D model from a 2D image?
Use a photo of one object on a plain background that fills at least half of the frame. Send it as image to an image to 3D model, such as flymyai/tripo-h3_1-image-to-3d, and download the model file from the answer. From Claude Code, Claude, ChatGPT or Cursor, connect the FlyMy.AI MCP and ask: turn this photo into a 3D model with Tripo H3.1. From code, send the photo to the model's predict/async route with your FlyMy.AI API key.
Is there a Tripo, Meshy or Hunyuan 3D API without a separate account?
Yes. On FlyMy.AI each of these models is an endpoint, such as flymyai/tripo-h3_1-image-to-3d or flymyai/meshy-7_1-image-to-3d, that you call with your FlyMy.AI API key or through the FlyMy.AI MCP. You need no Tripo, Meshy or Tencent account or key: each run is billed to your FlyMy.AI balance.
What file formats do the 3D models return?
Most generators return a GLB file (binary glTF) and a preview render. Tripo returns FBX instead of GLB when you ask for quad topology, Meshy adds an FBX copy of every model, and Hunyuan 3D 3.1 Rapid returns an OBJ file with its material file and a PNG texture. Meshy Rigging returns the rigged character as GLB and FBX, with walking and running clips.
How much does AI 3D model generation cost?
From $0.10 to $1.52 per model on FlyMy.AI, by model and settings. With the default settings, Tripo H3.1 costs $0.30 from a photo and $0.20 from a prompt, Hunyuan 3D 3.1 Rapid $0.225, Hunyuan 3D 3.1 Pro $0.375, Meshy T2.5 $0.60, Tripo P2 $1.10 and Meshy 7.1 $1.20. Rigging a character costs $0.20. There is no subscription or minimum.
Can I rig and animate the 3D model?
Yes, for humanoid characters. Meshy Rigging adds a skeleton to a GLB from any generator, with walking and running clips, for $0.20, or $0.32 with one animation from Meshy's library. Meshy 7.1 and Meshy T2.5 can rig in the same request with enable_rigging (+$0.20) and enable_animation (+$0.12). The skeleton can follow Mixamo, the UE5 mannequin, Unity humanoid, Maya HumanIK, Blender Rigify, 3ds Max Biped, CC5, Daz Genesis 9 or Meshy's own.
Can I use the models in Blender, Unity or Unreal Engine?
Yes. GLB imports into Blender and Unreal Engine, and FBX into Unity, Unreal Engine and Blender. Meshy returns both formats, Tripo returns GLB (FBX with quad topology), and Meshy's rigs can use the skeleton your tool expects, such as ue5_mannequin, unity_humanoid or blender_rigify_basic_human.
Can the same API make music and sound effects?
Yes. Mureka V9.5 Song sings your lyrics or writes its own ($0.15 with lyrics, $0.50 from a prompt alone), Mureka V9.5 Instrumental composes background music without vocals ($0.15), Lyria 3.5 composes a song or an instrumental from a description ($0.10), and Mirelo SFX 2.0 makes sound effects and ambience at $0.015 per second.
How long does an AI 3D model take?
Minutes, not seconds. In our tests Tripo H3.1 took 112 seconds from a photo and 203 seconds from a prompt without textures, and Meshy T2.5 took 80 seconds for an untextured preview mesh. Send long runs to the predict/async route and read the result when it is ready.
Make your first 3D model from a photo
Open Tripo H3.1 Image to 3D, upload a photo of one object and run it, or ask your assistant through the FlyMy.AI MCP. $0.30 a model with the defaults.