WHAT NEW IN X10.29 THU

AI collections from X · 22 collections · 60,637 posts

claudeVIDEO

Opus 5.5 directs a full brand trailer in Dreamfort

The author used Opus 5.5 as a director agent to storyboard, generate first and last frames with ChatGPT Images 2.5, write timed Seedance 2.5 prompts, drive browser rendering, and edit a full brand trailer.

Advait
@advait_jayantArchived 2026.09.26Originally posted 2026.09.25
Advait @advait_jayantOriginally posted 2026.09.25
Open original on X

Let me let you guys in on a little secret... I made a full-length brand trailer with Claude Opus 5.5 + ChatGPT Images 2.5 + Seedance 2.5 in Dreamfort today... and you can too! The future is already here: - I no longer need to spend hours prompting video models to generate a single scene. - Opus 5.5 can act as a director and can prompt these models in a great way for me. - I just give it notes, the way a director gives notes to a crew. My exact workflow: - Opus 5.5 acts as the director agent. - I first make a storyboard with Opus once I've sparred with it and finalized a plot. - Each scene in the storyboard is represented as a series of clips. - For each clip, I ask Opus to generate multiple reference images with ChatGPT Images 2.5: the first frame, the last frame, and a sheet for every character in it. - I give each character a name. Opus then ensures that each character is consistent across every scene. - Opus writes a timed Seedance 2.5 prompt for every clip: what each image is for, what happens second by second, every line of dialogue, every sound. - It even drives DreamFort in Chrome for me: uploads the frames, pastes the prompts, renders each clip. - Then it watches every take frame by frame and transcribes every line. When a shot is wrong, it redraws the keyframe instead of rewriting the prompt. - Finally it cuts the film, strips the model's own music, lays one score underneath, and upscales it to 4K. Here is the first film I made with this workflow:

x.com/advait_jayant/status/2103204555429216517Archived 2026.09.26 02:20