How our agency's AI process became Overs

Before Overs existed, we wired our AI photo jobs by hand as canvases in Figma Weave, the node-based AI workspace Figma added when it acquired Weavy in October 2025. The canvases always had three zones. Inputs sat on the left: product photos, reference pictures and brand notes. In the middle sat a locked "brain" of AI roles, each with fixed instructions. On the right, the output fanned out to about ten shots.

We built four campaigns that way in 2026. Here is what each one took.

Somebody looked at all 361 renders and threw away 316 of them. The spread between jobs was wide. Packaged gummies and trousers in a studio took 3 or 4 tries per keeper. The brownie ads took 12, and the surf job, one model carried from a studio to a cliff to the ocean, took 13.6.

The model fees were the small part. At the providers' published prices in September 2026, 8 renders cost about $0.27 on Nano Banana 2 Lite ($0.0336 a render, per Google's Gemini API pricing) and about $1.69 on GPT Image 2 at high quality ($0.211 a render, per OpenAI's image generation guide). The expensive part was the looking.

Lock the person before anything else

The surf job's character sheet, one model shown from several angles in the exact clothes, ran 20 times before we kept one. The first attempt had the longest prompt and 4 reference images, with a full list of what had to stay identical: face, hair, every seam and logo placement, body proportions. We didn't keep it. The next two were one-line prompts, and later attempts piled on up to 8 references. The sheet we kept, attempt 20, used 5 reference images and this prompt:

combine images provided to create a character sheet with various angles,
his face, body, and clothing must remain absolutely integral and consistent
show closeups for face
NO words on the image

The kept character sheet prompt from the surf job, as written.

Fewer, better-chosen references and a short prompt beat the longest prompt and the most references. The last line is there because character sheets tend to come back with labels printed on them.

Send each photo only the references it needs

On the trousers job, every store-page prompt named its pictures by number. INPUT IMAGE 1 was the approved character sheet and INPUT IMAGE 2 was the flat lay of the trousers. Those prompts ran about 330 words on GPT Image 2. The backdrop and lighting sentences were reused word for word across the store photos. Only the framing and pose changed from photo to photo, and the camera sentence changed once, for the fabric close-up, a 100mm macro at f5.6.

Kept prompts share a shape

The prompts behind kept photos ran about 140 to 200 words on Nano Banana models and about 300 on GPT Image. The Nano Banana ones followed the same order. They opened with the subject and a pointer to the reference image, named the background with a hex color, described light as something falling on a surface, listed the palette as names plus hex codes, gave one sentence of camera and lens, and ended with a separate sentence about grain.

The fixes lived in one person's habits

On the brownie job, all ten kept prompts asked for heavy film grain, and nine carried the same sentence: "Heavy visible film grain throughout." A person added it by hand, prompt by prompt. Only 14 of the job's 120 prompts carry it. Nothing on the canvas remembered that sentence, so somebody had to.

Overs was built on that process. It runs ten steps in order, the way a real shoot runs: research, brand rules, visual style, art direction, product analysis, casting, shot planning, prompt writing, a quality check, then the render. You can watch each step happen. Here is each hand-built habit next to the Overs step that does the same job.

Today 100 Creatives makes its campaign and product images in Overs. Abhi makes lots of images with it, across all of his businesses, and he credits Overs for the growth of 100 Creatives and his other brands.

Overs makes the stills. It doesn't make video and it doesn't write ad copy. Our video ads come from AI video tools, art-directed by our team, and Overs can hand over a text motion prompt for each photo, written for a video model such as Seedance or Veo.

The four campaigns above were built before Overs existed, so none of them was made in it. Client work is under NDA, so this page names no clients and no client campaigns.

Overs adds no markup to the AI. You connect your own OpenRouter key, a service that gives you one account and one bill for many AI models, and pay the image model directly. OpenRouter passes provider prices through without markup and charges 5.5% ($0.80 minimum) when you buy credits by card.

Take a 30-photo set: 10 packshots on white, 10 lifestyle scenes and 10 photos on a model. On Overs it comes to about $36 to $118. The low end assumes 3 tries per photo on Nano Banana 2 at $0.067 a render; the high end assumes 14 tries per photo on GPT Image 2 at high quality, $0.211 a render. Both include the Pro plan at $29, because 90 or more renders is past the free plan's 40 photos a month, and about $0.20 of planning for each of three runs, before OpenRouter's card fee.

The same set costs $2,026 at soona's published studio prices and $13,790 with a hired crew for one day, before usage fees, according to Overs' guide to what a product photo shoot costs. The high end of the Overs range is under 6% of the studio price.

What that leaves out is time. At 8 tries a keeper, someone looks at about 240 images to pick those 30, and some shots still need a camera. Our page on when you still need a real photo shoot covers which ones.

• It doesn't make video. It writes a text motion prompt for each photo, and a video model makes the clip.
• It doesn't publish to Instagram, ad accounts or stores. You export the files.
• It has no public API or MCP.
• It doesn't train a custom model on your product. It works from your reference photos.
• You can't try it without an OpenRouter key and a little credit, because the AI runs on your own key.

When a brand needs video, finished ads with copy and layout, or someone to run the whole job, that's agency work. 100 Creatives quotes it in writing from $4,000, and the breakdown of what an agency fee covers shows where that money goes.

Give Overs one photo of your product and your website. It studies the brand, plans the shots, and shows you the plan and the estimated cost before a single photo renders. For anything worn or held, you approve a character sheet first, so the same model appears in every photo.

The free plan makes 40 photos a month for one brand. Pro is $29 a month for unlimited photos and 3 brands, and Team is $99 a month for 10 brands and 5 seats (Overs pricing page, checked September 2026). The AI costs a few cents a photo on your own key.

Overs is our sister product, made by 100 Creatives' founder. Try Overs free at www.overs.studio, starting with one product and one set.