Corespeed Art — Multi-Model AI Media via fal.ai#

Auth: Set FAL_KEY with your fal.ai API key (get one at https://fal.ai/dashboard/keys).

Workflow#

1. Pick a model from the tables below
2. Read its reference file to get the exact endpoint and parameters
3. Run the command with the endpoint and JSON parameters

Usage#

bash

uv run {baseDir}/scripts/fal.py ENDPOINT --json '{"param":"value"}' -f output.ext [-i input.ext]

• ENDPOINT — the fal.ai model path from the reference file (e.g. fal-ai/nano-banana-2)
• --json — model parameters as JSON object
• -f — output filename
• -i — input file(s) to upload (repeat for multiple), auto-injected as image_url/ image_urls/ start_image_url/ video_url
• --audio — audio input file (for lipsync)

Image Generation#

ModelBest ForReferenceNano Banana 2Pro quality, web search, thinkingRead nanobanana.mdFLUX 2 ProPhotorealistic, zero-configRead flux.mdFLUX Schnell⚡ Fastest iterationRead flux.mdFLUX Pro v1.1Accelerated, commercial useRead flux.mdFLUX.1 Dev12B params, fine-tuning friendlyRead flux.mdGPT Image 1.5Transparent bg, instruction followingRead gpt.mdQwen Image 2 ProChinese+English, typography, native 2KRead qwen.mdRecraft V4 ProDesign/marketing, color controlRead recraft.mdSeedream 5 LiteMulti-image editing, reasoningRead seedream.md

Video Generation#

ModelBest ForReferenceKling v3 Pro I2VBest I2V, multi-shot, audio, 3–15sRead kling.mdSora 2 T2VLong video up to 20s, charactersRead sora.mdSora 2 I2VImage→video with SoraRead sora.mdVeo 3.1 T2VCinematic + native audio/dialogueRead veo.mdVeo 3.1 I2VImage→video with audioRead veo.mdLTX 2.3 T2V Fast⚡ Fast, up to 2160p/20s, open sourceRead ltx.mdLTX 2.3 I2VImage→video, start+end frameRead ltx.mdPixverse v5 I2VAnime, 3D, clay, cyberpunk stylesRead pixverse.md

Audio / TTS#

ModelBest ForReferenceMiniMax Speech-02 HD30+ languages, loudness normalizationRead minimax-speech.md

Music & Sound Effects#

ModelBest ForReferenceBeatoven MusicAI music, up to 90sRead beatoven-music.md

Utilities#

ToolBest ForReferenceTopaz UpscaleAI image/video upscale 2x–4xRead topaz.mdBRIA RMBGProfessional background removalRead bria-rmbg.mdSync LipsyncAudio-driven lip sync on videoRead sync-lipsync.md

Notes#

• No manual Python setup required. The script uses PEP 723 inline metadata. uv run automatically creates an isolated virtual environment and installs the fal-client dependency on first run.
• fal.ai uses a queue system — the script polls until generation completes.
• Video generation can take 30s–3min.
• Use timestamps in filenames: yyyy-mm-dd-hh-mm-ss-name.ext.
• Script prints MEDIA: line for OpenClaw to auto-attach.
• Do not read generated media back; report the saved path only.