ChatGPT 6 Astra Just Changed Content Forever
What this video is
⚡ a 15-minute video, readable in 60 secondsA creator on a $100/month ChatGPT Pro plan connects Codex (running GPT-6 Astra Ultra) to DaVinci Resolve's paid MCP server to test whether AI can build finished animations and edit full videos by directly controlling the editing software rather than just describing steps. He runs three animation experiments of increasing complexity (a minimalist character scene, a purple 'step' motion graphic, and a 12-second ship battle), then a bigger test: auto-editing a full 12-minute talking-head video with all graphics generated through Higgsfield. He tracks real time and credit cost at each stage and argues this workflow is early evidence of a coming shift in how video editors work, while noting limits like the paid-DaVinci requirement and rough spots like broken water-impact effects.
[00:46] Starts on the ChatGPT Pro plan ($100/month) with a $0 credit balance
Key takeaways
- [00:59] Uses Codex with the model set to GPT-6 Astra Ultra at 'ultra' effort
- [01:15] Test 1 prompt: a 15-second, 1920x1080 flat 2D animation in Iman Gazi's style, minimalist person at a computer with a city skyline, in deep navy, royal blue, and icy blue
- [03:15] Follow-up prompt adds motion: the character typing or moving the mouse, lights on the city buildings, and plant movement, to make the scene feel more dynamic
- [03:53] The whole first animation (initial render plus motion revision) took about 20 minutes
+ 13 more takeaways
- [04:34] Test 2 prompt: a purple-gradient DaVinci Resolve motion graphic like an After Effects animation, with a curved line hitting three dots that each explode into 'step one / step two / step three' text
- [05:32] Watching the underlying Fusion node graph, he says 'a human can't make this'
- [06:13] Test 3: two warships firing at each other on a blue gradient sea, using 2D illustrations with a 3D camera for zoom and parallax, specified as a roughly 12-second 'mini war animation'
- [06:44] The battle animation took 24 minutes total to complete, finishing around 7:13 p.m.
- [07:00] Calls the result impressive and hard to tell is AI, despite the bullets/water impacts looking wrong
- [08:35] Clarifies ChatGPT isn't just prompting a program, it's actually controlling the computer
- [08:44] This entire workflow only works with the paid version of DaVinci Resolve
- [08:53] Plans a bigger test: ChatGPT and Claude editing a full 12-minute talking-head video with no existing graphics, generating everything via Higgsfield
- [11:41] That full 12-minute video edit took about 30 minutes of real time
- [12:01] Only 27 Codex credits remained afterward
- [12:12] Roughly 70 Higgsfield images were generated for it at about 4 credits each, an estimated 300-500 Higgsfield credits total
- [12:55] Calls it his best-performing YouTube upload so far, earning YouTube's fireworks animation in Studio
- [14:12] Warns that just listing events with no tension or conflict makes a video predictable, calling it the 'And-Then Trap'
Shown on screen — grab and go
PROMPTThe Blue Hour apartment animation prompt
Create a finished 15-second, 1920x1080 animation [...] soft shadows, and restrained glows. Show a minimalist person working at a computer inside an apartment. Behind them, a large window reveals a stylized city skyline. Construct everything from flat 2D shapes—the person, desk, monitor, window, and buildings. Keep the person anonymous and very simple, with no facial details. Let the composition, gradients, lighting, and shadows create the visual richness. Use deep navy, royal blue, and icy blue highlights. Add a soft blue glow from the monitor, gentle window light, and subtle shadows between overlapping shapes. Separate the flat artwork into depth layers. Use a slow, smooth camera pullback to reveal more of the room and the city, with gentle parallax between the layers. The artwork must retain its flat, illustrated cutout appearance throughout. Keep clear space at the top for the exact headline: Subscribe to Danny Why. Use bold white text with a tasteful soft glow. Fade it in near the beginning and keep it fixed and readable. Ease the camera into a clean final hold. Make reasonable creative decisions, preview and refine the result, preserve existing work, and deliver both the editable composition and finished video
Merged from the same prompt shown scrolled across the chat panel at 01:15, 01:31, and 01:35; one short word-gap marked [...] near the start. shown at 1:31
PROMPTThe purple curved-line step-counter prompt
Use DaVinci to create a motion graphic where the background is slightly purple and has those beautiful gradients, purple, and everything is beautiful, like an After Effects animation. And you're basically going to create this beautiful curved line that starts from bottom left and goes to the bottom right, but it goes in a beautiful curves, and it hits three dots. And every time it hits a dot, the dot explodes and says step number one, step number two, step number three. But it says st[...] two, st[...] three with a very beautif[...] font with a beauti[...] glow on it, beautiful animation. Everything is very smooth as if it's some professional YouTube video
Merged from two scrolled reads of the same chat message at 04:34 and 04:36; a few short gaps marked [...] where OCR dropped characters. shown at 4:34
PROMPTThe add-motion-to-the-scene follow-up prompt
[...] some lights to the blocks in the city, the buildings in the city. Add some light to the city buildings. Add some movement to the plant. Just make it feel a little bit more dynamic.
Transcribed from the on-screen chat message at 03:17-03:19; the opening word is cut off at the frame edge, marked [...]. shown at 3:17
Their links, sorted & clickable
🔗 Other links1
👉 Get the Higgsfield SkillWas this brief useful?