# Idol Max vs Turbo on fal: Pricing, Rigging and Which GLB to Import

Choose between Idol Max and Turbo, understand their GLB outputs, and check a generated character before adding it to your game pipeline.

By Paras Madan | 2026-11-19


A generated character can look convincing in a preview and still arrive in your game without a skeleton. With Idol Max, picking the right output file matters as much as picking the model.

fal's new Max and Turbo endpoints turn an image into a textured game mesh, with optional rigging and animation. That is a different workflow from the original Idol splat endpoint. The useful question now is which asset to request, which file to import and what to check before generating a whole collection.

## Choose Max or Turbo by the texture you need

fal describes Max as the quality option: its base colour is repainted from the input image. Turbo projects colour from the reconstructed shape instead. Both return a game mesh with roughly 30,000 triangles, UVs, a 2K texture and a normal map. [fal's Idol Max overview](https://fal.ai/idol-max) explains the distinction.

For a prototype, I would start with Turbo and compare it against Max on the same reference image. Use something representative of the actual project, not a deliberately easy demo object. A small background prop and a character seen in close-up deserve different acceptance criteria.

Judge the comparison at the camera distance players will see. A difference that disappears at gameplay scale may not justify changing the pipeline. A face, emblem or costume detail that becomes unreadable probably does.

## Budget for the variant, not just the model name

The listed per-asset prices are $0.20 for Max and $0.10 for Turbo. With rigging and animations, they become $0.25 and $0.125 respectively. At those published rates, 100 rigged assets would be $25 versus $12.50 before other costs.

These are generation budgets, not approved-asset budgets. Track how many outputs survive your review and how much cleanup each needs. If a cheaper variant produces twice as much repair work, its lower API bill may be the least important number in the comparison.

## Request one clear subject

The [Max API schema](https://fal.ai/models/speridlabs/idol/max/api) requires `image_url`. Set `image_type` to `real` for photographs or `synthetic` for renders and generated images. Rigging is intended for characters and creatures; the documentation warns that rigid props can receive unusable skeletons.

For an initial evaluation, I would use a character with clearly separated limbs and an unobstructed silhouette. Keep that reference unchanged while comparing variants. Changing the image, model and options together makes it harder to understand why one result looks better.

Save the input beside the output in your own asset record. Give the record a short review note such as “good silhouette, weak shoulder deformation” rather than an unqualified pass.

## Import the output that matches your job

The response distinguishes three files:

- `game_glb`: the textured game mesh, without a skeleton.
- `rigged_character_glb`: the mesh with skeleton, skin weights and compatible requested clips, when rigging succeeds.
- `model_glb`: the full-resolution, vertex-coloured mesh.

Check `rig.succeeded` and the actual `rig.animations` list. Unsupported clips can be omitted. Also, `mesh_stats.num_faces` describes the full-resolution mesh, not the smaller game asset.

My suggested import checklist is deliberately separate from the API's success status: confirm the chosen file opens in your target tool, inspect materials under your scene lighting, and play every returned clip. Watch shoulders, knees and feet through the full loop rather than judging the first frame.

## Treat shading as a separate decision

The `enable_pbr` option adds roughness and metallic maps without changing base colour. fal says this stage typically adds 20 seconds to two minutes.

I would leave it out of the first geometry comparison, then evaluate it on whichever asset survives. Use the same scene and lights for both versions. Otherwise a lighting change can make a material upgrade appear better or worse than it is.

Do not assume a good preview means the asset fits your performance budget. Check the imported asset in the intended scene before placing dozens of copies.

## Keep generation separate from the page request

fal's [queue documentation](https://fal.ai/docs/documentation/model-apis/inference/queue) recommends asynchronous inference. Submit a job, retain its request ID, then poll for status or receive the result through a webhook. The lifecycle distinguishes waiting in the queue from active processing and completion.

For an asset tool, I would show generation and review as separate states. “Completed” should mean a result is available; “approved” should mean someone checked the actual asset. That small distinction prevents an automated batch from silently becoming a production asset library.

I'm Paras. I work at the intersection of AI, engineering and distribution.

I help technical teams turn API details into useful documentation and product storytelling that developers can act on.

[Book a 30-minute call](https://cal.com/paras-madan/30min).
