An AI video trailer in about one day
Robert Lester
Creative director Robert Lester turned his essay 'Art in the Age of AI' into a video trailer in about a day, chaining a different AI tool for each job: images, upscaling, animation, voiceover, music and the final edit.
Story written by: the person who did it
The problem
Robert Lester had written an essay about art and AI and wanted strong images for it, then decided to turn those images into a video trailer. AI image tools only make small, low resolution files, and traditional animation is slow. He wanted to see how fast a chain of AI tools could produce a finished trailer.
How it was built
- 1Plan images from the essayPerson
- 2Generate the editorial imagesComfyUI
- 3Upscale the imagesMagnific.ai
- 4Storyboard the trailerPerson
- 5Animate the shotsDream Machine
- 6Record the voiceoverDescript
- 7Make the soundtrackSuno
- 8Edit the trailerPremiere Pro
Tool step Person step
1. Plan images from the essay
Lester read each section of his finished essay, pulled out the key themes and sketched visual concepts and prompts for each one, such as reworking Leonardo's Vitruvian Man. Planning and image generation together took about 6 to 8 hours (self-reported).
Check: He checked that every image idea came from a real theme in that section of the essay.
2. Generate the editorial images
He built a ComfyUI workflow where one image made in Midjourney sets the composition and another sets the style next to the text prompt, and he used Photoshop to adjust images. He reused his own earlier images as style references to keep the look consistent.
ToolComfyUI
Check: He checked new images against his earlier style references so the set stayed consistent.
3. Upscale the images
Because AI generators output images around 1024 x 1024 pixels, he ran them through creative upscaling in Magnific.ai, which regenerates the image in tiles at a higher resolution and adds detail guided by a text prompt.
ToolMagnific.ai
Check: He compared the before and after versions of each upscaled image.
4. Storyboard the trailer
He rethought the article images as shots for a trailer, made extra images to fill gaps and replaced weak ones that would look awkward in a video. This took about 2 hours (self-reported).
Check: He checked the storyboard had no gaps and a coherent style across the shots.
5. Animate the shots
He fed the storyboard frames into LumaLab's Dream Machine as start and end frames, sometimes adding prompts to guide the movement. Each shot needed several attempts, about 4 hours in total (self-reported).
ToolDream Machine
Check: He reran each shot until the movement worked for that shot.
6. Record the voiceover
He wrote a voiceover script, recorded and edited it in Descript, and used Descript to render animated captions. This phase took about 3 hours (self-reported).
ToolDescript
7. Make the soundtrack
He generated the soundtrack with Suno using the prompt 'pizzicato strings and subtle deep glitch drums soundtrack for a video essay, soothing yet ominous', and after 10 generations he chose the first one it had created.
ToolSuno
Check: He listened through the 10 generations before choosing one.
8. Edit the trailer
He laid out all the media in Premiere Pro and cut a tight final edit, which finished the trailer.
ToolPremiere Pro
Before and after
Before
AI image generators output images around 1024 x 1024 pixels, about a quarter of 4K resolution.
After
Creative upscaling in Magnific.ai regenerated each image in tiles at a higher resolution, with extra detail added by a new guided generation pass. source
What made it work
- He planned every image from the essay's themes before generating anything.
- He reused his own earlier images as style references to keep a consistent look.
- He used LumaLab's Dream Machine as a, in his words, brute force animator, much faster than a traditional animation process.
- He accepted iteration: several attempts per animated shot and 10 soundtrack generations.
- He gave each job its own tool instead of forcing one tool to do everything.
Follow along
Practice steps suggested by Ai Siftly, based on the account above. We have not run them ourselves.
Choose six things the viewer should understand
Example to copy: source paragraph | idea | visible scene | why it belongs
You should get: Six-row storyboard.
Check: No new factual claim introduced.
Make six frames
Example to copy: Visual scene: [beat]. Consistent style: [description]. Avoid text/logos/real-person likeness.
You should get: Six still frames.
Check: Rights and visual consistency checked; generation optional.
Write narration
Example to copy: Write a twenty-second narration from these checked beats, without adding facts: [beats].
You should get: Timed narration script.
Check: Every claim traces to the article.
Arrange the pictures in order
You should get: A timed still-frame sequence.
Check: Beginning and conclusion are clear.
Review the storyboard or save the video
You should get: A twenty-second storyboard and script.
Check: Inspect the actual output.
Try it yourself
Find a matching workflow in the workflow finder. Search words: a r t i c l e t o A I v i d e o t r a i l e r.
Sources and notes
- Art in the Age of AI | Visual Case Study (Robert Lester Creative, )
- First generation setup (ComfyUI, )
- Image-to-image tutorial (ComfyUI, )
- All timings and the 'one day' claim are self-reported by Robert Lester.
- The source reports no audience or viewing results for the trailer.
- No date appears on the source page, and tool behaviour may have changed since it was written.
- The source review did not establish a downloadable workflow file or node list, so the follow-along guide does not recreate his exact ComfyUI setup; current ComfyUI setup needs models, hardware and licence checks.
Independent first-person. Page date not found, so date is null. Historical tool behaviour and commands are as documented at the source date; current versions may differ. Record dated 2026-10-27.