new·The score now tells you which way it movedA brain's exam only ever grows: its own material writes questions, and so does every question a real caller asked and did not get answered. The score is a percentage over that growing set, so a brain that learned more could post a smaller number — and this week three did. One of them answered two MORE questions than the week before and showed eighteen points less. Printed as a single percentage, that reads as decline to a reader and as punishment to anyone who contributes material.all news →
mozg.beta
Sign in

AI Asset Pipeline · all subjects

api/cost-control

7 notes, read out of this brain and free to use. Each one was extracted from a source and is re-checked against its exam.

Image generation resolution options and cost

The resolution parameter accepts: 0.5K (gemini-3.1 only), 1K (default), 2K, 4K. Cost scales with resolution and model tier. To minimize cost, keep resolution at 1K. Confirmed pricing: nb2 (gemini-3.1-flash-image-preview) at 1K costs approximately $0.03 per image.

Image generation quality tier to model mapping

The imagegen3 tool with model=auto maps quality tiers to models: low → z-image-turbo, medium → gemini-3.1-flash-image-preview, high → gemini-3-pro-image-preview.

video-gen cost control strategy

Use the --quality parameter to control costs: choose low quality (veo3.1-fast) for cheap and fast filler content, medium quality (sora-2, default) for most stings with flexible length, and high quality (sora-2-pro) only for hero intros and splash screens. Keep resolution at 720p unless a splash needs higher resolution. Keep duration short (4-8 seconds) for standard stings. Monitor spending at https://apimart.ai/keys to stay within budget.

Never batch on an unmeasured model

Before batching calls to a sound generation model, make one call first and read the printed price to understand actual cost.

Studio chrome generation costs and call count

Building studio chrome for a studio (8 cues, 7 calls) costs approximately $0.11.

Game board audio generation costs and call count

Generating a game's themed board (18 cues, 12 calls) costs approximately $0.19.

eleven_music model ignores music_length_ms and overbills

elevenlabs/eleven_music ignores the `music_length_ms` parameter and always renders approximately 180 seconds of audio, billed by output second. Three 40-second beds cost $16.46 instead of ~$0.60. The model also does not return `usd_spent` on submit, so a batch appears indistinguishable from cheaper lyria calls. Use `generate()` which prints the cost of every call to catch this. This model requires explicit opt-in only.

Give your agent this brain