Use fal on GPT-6 Astra to Create Videos & ImagesYour fal account, inside ChatGPT and Codex
GPT-6 Astra is OpenAI's most capable model to date, available in ChatGPT and Codex. Sign in with your fal account, or connect your fal API key, and GPT-6 Astra becomes a production-grade creative engine: it searches more than 1,000 AI video, image, audio and 3D models on fal, picks the right one, runs it, and hands the results back in the same conversation.
From fal's walkthrough: the CineLive app GPT-6 Astra built in Codex, streaming MiniMax H3 Max Director video in real time. Watch the full video.
GPT-6 Astra Community Demos & Examples
GPT-6 Astra's breakout demos are about building: 3D worlds, games, buildings, edits, brand systems, music. Every one of them needs finished media at some point, and that is where fal's models come in. One credited community demo per pattern, and the fal models that finish the job.
Astra builds the scene, fal supplies the assets
GPT-6 Astra writes Blender Python, places objects, assigns materials and lights scenes, then hands editable geometry to Unreal or the browser.
Finish it with fal: Meshes from a photo or sketch with Hunyuan 3D on the 3D models page, reference boards and textures from GPT Image 2.5, texture upscales with Topaz.
Community demo · Tom KrchaAn old steam-train drawing handed to GPT-6 Astra and rebuilt as thousands of editable Blender objects.x.com/tomkrcha →Astra ships the game, fal makes it look and sound finished
From idea to assets, gameplay and a hosted prototype in one run. The graphics are the part that separates a demo from a game, and image generation is how the community is solving it.
Finish it with fal: Sprites, tiles and key art from GPT Image 2.5 or Nano Banana Pro, effects from Sonilo, a score from MiniMax Music 3, a trailer from MiniMax H3 Max.
Community demo · AnshuAn isometric 3D game one-shot in about 45 minutes, by the author's account, who says the trick to its graphics was image generation.x.com/anshuc →Astra models the building, fal renders the pitch
Floor plans, photos or a property listing go in; a Blender model with furniture, materials and lighting comes out, then an Unreal walkthrough.
Finish it with fal: Staged renders from GPT Image 2.5 or Seedream 5.0, walkthrough clips from a still with Kling 3, print-ready boards with Topaz.
Community demo · Yunfan YeA Zillow listing turned into a 3D model of the house from the listing photos, then a promotional video.x.com/realYunfanYe →Astra runs the edit, fal generates the footage
GPT-6 Astra assembles edits, animates meshes and drives DaVinci Resolve through computer use, the workflow in fal's own Resolve walkthrough below.
Finish it with fal: Shots from MiniMax H3 Max, Kling 3 and Seedance 2.5, sound design from Sonilo, a score from MiniMax Music 3.
Community demo · Duncan TrussellThe Backrooms built in Blender with a VHS treatment and sound effects; about 30 minutes to make, by the author's account.x.com/duncantrussell →Astra lays it out in Figma, fal draws the imagery
GPT-6 Astra turns visual references into editable Figma frames, decks and sites, and carries one direction through production tools.
Finish it with fal: The mockups in these demos come from GPT Image 2.5, the same model on fal with an API; campaign imagery and edits from Nano Banana Pro.
Community demo · @Design_emottoA corporate site drawn with GPT Image 2.5, rebuilt as Figma frames by GPT-6 Astra in Codex, then shipped; 30 to 40 minutes, per the post.x.com/Design_emotto →Astra plays the studio, fal writes the tracks
GPT-6 Astra drives Ableton Live and FL Studio, writes notation and integrates audio into games.
Finish it with fal: Full tracks from MiniMax Music 3, voices from ElevenLabs, effects from Sonilo.
Community demo · timour kostersGPT-6 Astra driving Ableton Live from a one-line prompt to make a track.x.com/timourxyz →Community demos are public posts by their authors about GPT-6 Astra, credited and linked; the demo clips shown are the authors' own, trimmed for the tile. fal was not involved in the demos, and the claims are the authors' own as posted. Links verified September 15, 2026.
How do you connect fal to GPT-6 Astra?
Three ways in, one fal API key. Pick the surface you already use GPT-6 Astra in: the fal plugin for ChatGPT, the fal MCP server, or the fal CLI inside Codex.
1. Get a fal account and API key
The fal plugins for ChatGPT sign you in with your fal account, so a free account is all they need. The MCP server in other clients and the fal CLI use an API key from the dashboard, created in a few seconds. Either way, every generation GPT-6 Astra triggers is billed to your fal account at standard fal pricing.
Get your API key →2. Give GPT-6 Astra access to fal
The fastest route. Two fal plugins are listed in ChatGPT: Fal (Generate with Fal) and Fal MCP. Install Fal and sign in with your fal account; nothing to paste.
- 01Open Plugins in ChatGPT, search for fal, and add Fal (Generate with Fal). Add Fal MCP too if you want the raw MCP tools.
- 02Sign in to fal.ai when prompted, with GitHub, Google or your enterprise email. The plugin uses your fal account, so there is no key to paste.
- 03Start a chat on GPT-6 Astra, pick the Fal chip in the composer or just ask, and describe what you want. The plugin picks the model and the file comes back in the chat with a download link.
fal in ChatGPT
Search, run and chain more than 1,000 image, video, audio and 3D models on fal without leaving the conversation. Sign in with your fal account; generations are billed to it at standard fal pricing and the plugin itself is free.
chatgpt.com/plugins/fal3. Ask GPT-6 Astra for anything
GPT-6 Astra searches the fal catalog, checks the price, picks a video model with native audio, runs it, pulls the stills and hands back the files. You never choose an endpoint or read a schema.
Install the fal plugin
One click in the ChatGPT plugin directory, then sign in with your fal account. GPT-6 Astra can call every model on fal.
Manage your API keys
Create, rotate and revoke the keys GPT-6 Astra uses. Every run shows up in your fal usage with its cost.
MCP and CLI reference
Setup guides for every client, the full tool reference, and the fal api command for the terminal.
What does GPT-6 Astra look like working with fal?
Two walkthroughs from fal's YouTube channel, recorded in the first week of GPT-6 Astra. One builds a live AI video app in Codex; the other runs GPT-6 Astra inside DaVinci Resolve with fal's H3 Max plugin.
GPT-6 Astra | How to build an AI Live Stream App with H3 Max Director
5:16 · fal · September 3, 2026
To test GPT-6 Astra as a coding model, fal builds an AI live-stream app that takes direction from chat. GPT-6 Astra writes the app in Codex; MiniMax H3 Max Director on fal streams the video in real time and takes live direction from viewers.
DaVinci Resolve 21.1 plus GPT-6 Astra is Editing AGI!
14:06 · fal · September 8, 2026
Blackmagic Design's Resolve 21.1 ships new AI Assistant features. fal drives a full edit with GPT-6 Astra, from an H3 Max reference-to-video plugin through assembly, audio mixing, color grading and animated subtitles.
What can GPT-6 Astra create with fal?
Six briefs for one invented espresso brand, Aurelio. Every asset below was generated through fal's endpoints on September 14, 2026: the brief, the output, the models that ran and what the run cost.
Product and campaign video
MiniMax H3 Max image to video, 10 seconds at 1080p with native audio, starting from the product photo so the machine stays the same one.
Posters, packaging and product shots
GPT Image 2.5 Flare for the poster and its type, then Topaz Precision at 3x for a 3600 x 4800 print file.
Voiceover and soundtrack
ElevenLabs Eleven v3 reads the 259-character script; MiniMax Music 3 writes the instrumental. Mixed and loudness-normalized in one turn.
Bring a photo to life
Kling 3 Standard image to video, 5 seconds, audio off. GPT-6 Astra uploads the photo to fal first with upload_file.
Assets for games and AR
Hunyuan 3D 3.1 Pro image to 3D, textured GLB from one photo. Drag to orbit it. More on the 3D models page.
Concept to deliverables in one conversation
Six frames from GPT Image 2.5, two animated with MiniMax H3 Max image to video, scored by MiniMax Music 3. Three labs, one conversation, one bill.
Briefs are what a user types; the model-level prompts GPT-6 Astra writes from them are longer. Costs are the prices published on each model page on September 15, 2026, billed to your fal account. MiniMax H3 Max is on a 50% launch discount through the end of September, so its cards show the rate charged today and the list rate that follows. MiniMax Music 3 is priced per track on its model page.
GPT-6 Astra can call every model on fal
All 1,000+ public models on fal, from the labs that build them, are available to GPT-6 Astra with the same API key. No allowlist, no per-model setup. A few popular starting points:
FLUX 3AI video
Kling 3AI video
Seedance 2.5AI video
MiniMax H3 MaxAI video
Wan 3AI video
Veo 3.1AI video
LTX 2.5AI video
GPT Image 2.5AI images
Nano Banana ProAI images
Seedream 5.0AI images
FLUX.2AI images
TopazUpscaling
ElevenLabsVoice and music
MiniMax Music 3Voice and music
3D models3D- 1,000+ moreEvery public model on fal, searchable by task. GPT-6 Astra can reach all of them.Explore all models →
Why is content from GPT-6 Astra on fal production-grade?
Because nothing between the prompt and the file is a toy. GPT-6 Astra calls fal's production endpoints with your key, and the result is the same asset an engineer would get from the API.
The same models teams ship with
fal serves the production API for FLUX, Kling, Seedance, MiniMax, GPT Image and more than 1,000 other endpoints. GPT-6 Astra calls those exact endpoints, so what you get in chat is what you get from the API.
Your key, your billing
Every run is billed to your fal account at standard fal pricing, with no markup for using it from GPT-6 Astra, whether you signed in through the plugin or connected a key. GPT-6 Astra can check a model's price with fal's pricing tool before it generates.
Files, not previews
Outputs come back as URLs on fal's CDN at the resolution you asked for: MP4, PNG, WAV, GLB. Download them, share them, or pass them into the next step.
Stateless by design
The fal MCP server keeps no sessions and no state. Your API key travels per request in the Authorization header and is never stored, and you can rotate or revoke it from the dashboard at any time.
Information updated as of September 15, 2026.
What does GPT-6 Astra do with fal's tools?
fal exposes eleven tools over MCP. GPT-6 Astra picks them on its own: you describe the result, it searches, prices, runs, uploads and polls.
search_modelsSearch 1,000+ models by keyword or category.
recommend_modelDescribe the result and get model recommendations.
get_model_schemaRead every input and output parameter for a model.
get_pricingCheck the cost per run or per second before generating.
search_docsSearch fal documentation for guides and examples.
run_modelRun a model and wait for the result.
submit_jobQueue long jobs: video, 3D, training.
check_jobPoll a queued job for status and logs.
get_job_resultFetch the output of a finished job.
cancel_jobCancel a job you no longer need.
upload_fileUpload a photo, clip or sketch to fal's CDN as model input.
Common questions about fal on GPT-6 Astra
What is GPT-6 Astra?
GPT-6 Astra is OpenAI's newest and most capable model, released on September 3 and 4, 2026. It is available in ChatGPT on the Plus, Pro, Business and Enterprise plans, in Codex, and through the OpenAI API as gpt-6-astra. It stands out for computer use, coding and long multi-step tasks, which is exactly what makes it a strong operator for the models on fal.
Can GPT-6 Astra generate videos and images?
Yes, through fal. Sign in through the fal plugin for ChatGPT, or connect your fal API key through the fal MCP server or the fal CLI, and GPT-6 Astra can run more than 1,000 AI image, video, audio and 3D models on fal, including FLUX 3, Kling 3, Seedance 2.5, MiniMax H3 Max, GPT Image 2.5 and Nano Banana Pro. It picks the model, runs it and returns the files in the conversation.
Do I need a fal account to use fal on GPT-6 Astra?
Yes. Create a free account at fal.ai. The fal plugins for ChatGPT sign you in with that account (GitHub, Google or your enterprise email), so there is nothing to paste. For the fal MCP server in other clients and for the fal CLI, create an API key in the dashboard at fal.ai/dashboard/keys; the same key also works for the fal API.
What does it cost to use fal from GPT-6 Astra?
Connecting is free. The fal plugin, the fal MCP server and the fal CLI carry no charge of their own. You pay for the model runs GPT-6 Astra triggers, at the same standard fal pricing as calling the API directly, billed to your fal account. Your ChatGPT or Codex subscription is separate and billed by OpenAI.
Which ChatGPT plans can connect fal to GPT-6 Astra?
Both fal plugins, Fal and Fal MCP, install from the ChatGPT plugin directory at chatgpt.com/plugins/fal; on Team, Business and Enterprise workspaces an admin controls which plugins are allowed. Adding the MCP server yourself as a custom connector needs a paid ChatGPT plan with Developer mode turned on (Plus, Pro, Team, Enterprise or Edu). Codex CLI reads the fal MCP server from its own config, independent of ChatGPT settings.
Is my fal API key stored anywhere?
The ChatGPT plugins never handle your API key at all: they sign you in to fal.ai, and you can revoke that access from your fal account. When you connect the MCP server yourself, it is stateless: your key is sent per request in the Authorization header and is never stored on fal's side. Rotate or delete the key in the fal dashboard at any time and every connection that used it stops immediately.
Can I use fal with GPT-6 Astra in Codex CLI?
Yes. Register the fal MCP server once with codex mcp add fal-ai --url https://mcp.fal.ai/mcp --bearer-token-env-var FAL_KEY, then run codex -m gpt-6-astra and fal's tools are available in every session. Codex can also call the fal CLI directly: pip install fal, set FAL_KEY, and fal api <model_id> key=value runs any model on fal from the terminal.
What is the difference between the fal plugin and the fal MCP server?
ChatGPT lists two fal plugins. Fal (Generate with Fal) is the packaged experience: a Fal chip in the composer, the plugin picks the model, and the file comes back in the chat with a download link. Fal MCP is the hosted fal MCP server at mcp.fal.ai/mcp packaged for ChatGPT, exposing the eleven raw tools. Outside ChatGPT, any MCP client (Codex, Claude Code, Cursor, Windsurf, the ChatGPT desktop app) connects to that same server with the URL and your API key. All routes reach the same fal models.
Is GPT-6 Astra itself a model on fal?
This page is about using fal from GPT-6 Astra. GPT-6 Astra runs in ChatGPT, Codex and the OpenAI API. fal hosts OpenAI's generative media models, including GPT Image 2.5 and GPT Image 2, alongside more than 1,000 models from other labs, and GPT-6 Astra can call all of them.
Ready to transform your enterprise with AI?
Take the first step towards AI-driven innovation. Our team of ML engineers is ready to help you prototype, develop, and scale your AI solutions.