PARTLY

Partly true: 'Opus 5.5 VS GPT-6 Astra' is a real pair of models, but neither one made a frame of this video; Seedance 2.5 did, and the 'VS' is a prompt-writing contest with no rules shown

Facebook · Oct 2026

The claim'Opus 5.5 VS GPT-6 Astra.' Ten seconds of side-by-side cinematic clips (a wave, a castle, a cathedral on an island) labeled 'Opus 5.5' on top and 'GPT-6 Astra' below. Caption: 'Like this post and comment Video 83 to get access to the prompts. Tools: I2V: Seedance 2.5 by @dreamina_ai. Edit: Premiere Pro.'

Both names are real. Anthropic announced Claude Opus 5.5 on Sep 22, 2026, and GPT-6 Astra is listed on OpenAI's own models page. But the reel is not what the labels suggest. Anthropic's models page says every current Claude model takes text and image input and produces text output, and OpenAI's models page says the same for GPT-6 Astra: text out, no video, no images. Neither model can render a wave or a castle. The caption admits where the pictures came from: Seedance 2.5, ByteDance's video model, run as image-to-video inside Dreamina. So the honest reading is that each language model was asked to write a prompt (or a prompt plus a still), and the same video model rendered both. That is a comparison of prompt writing, filtered through one renderer, with no prompt, seed, or settings shown, so you cannot tell whether the differences come from the models or from Seedance's own randomness between runs. Which side 'won' is never stated either; it is a vibe clip. The sell is the comment-for-prompts hook ('comment Video 83'), which grows the page and feeds a DM list. There is no course visible, but the hook is the product. If you want the result, you do not need either language model: Dreamina sells Seedance 2.5 access on its own credit plans, and the prompts are the only thing being withheld.

What holds up

Reel read frame by frame: every frame is a cinematic clip with an 'Opus 5.5' or 'GPT-6 Astra' label; no prompt, settings, scoring or winner is ever shown. Caption credits 'I2V: Seedance 2.5 by @dreamina_ai' and Premiere Pro for the edit.
Anthropic news page: 'Introducing Claude Opus 5.5' dated Sep 22, 2026. Anthropic's models overview lists Opus 5.5 at $4 input / $20 output per million tokens and states all current models support 'text and image input, text output'. No video generation.
OpenAI's models page lists gpt-6-astra as a flagship model with text and image input and text output only; no video generation model is listed there. x-ref #306 for the audit's earlier Astra check.
Dreamina's site lists Seedance 2.5, Seedance 2.0 and 2.0 Mini as its video models, with free daily credits and paid plans; ByteDance's Seed page describes Seedance 2.5 as an audio-video model generating up to 30 seconds per run. That is the tool that produced the footage.

What doesn’t

The labels imply two AI models made two videos; in fact one video model (Seedance 2.5) rendered everything and the language models at most wrote the prompts.
No prompts, seeds or settings are shown, so differences between the rows cannot be attributed to Opus or Astra rather than to Seedance's run-to-run variance.
Comment-for-prompts hook: the only withheld item is the prompt text, used to grow the page and a DM list.

The catch

You are watching a Seedance 2.5 showreel wearing two famous model names. Opus 5.5 and GPT-6 Astra are real and expensive reasoning models, and neither outputs a single pixel of video; if you pay for them hoping to make clips like these you will be disappointed, and if you buy Dreamina credits you will get this footage without either.

How to actually do it

If you want these visuals, open Dreamina, pick Seedance 2.5 image-to-video, start on the free daily credits, and generate from your own still image plus a one-paragraph camera-and-mood prompt.
If you want a language model to write your prompts, test it yourself: give Opus 5.5 and GPT-6 Astra the same brief, run both prompts through the same Seedance settings and seed, and compare three runs each before you decide which prompt writer you prefer.
Skip the comment hook. Write down the shot (subject, camera move, light, lens) and you have the prompt the page is withholding.

The two model names check out and the footage is real AI video, but the reel's framing is wrong about who made it. Seedance 2.5 in Dreamina rendered both rows; the language models could only have written prompts, and no prompts or settings are shown. Anyone can reproduce the look on Dreamina's credits without the comment funnel.

Confidence: High
Posted by: an AI-video showcase page (about 84k views on this reel); reel sent to Buddy Tue Sep 29, 2026, ~7:58 AM

See the original claim →

We test hype for free. We build the real thing for a living.

Thirty minutes, no pitch — and you'll leave with something useful either way.

Book a call with Todd or start with the free Business Checkup →

The Verdict Weekly

Three verdicts every Friday. Free forever, unsubscribe anytime, no spam — that would be ironic.