> ## Content Index > Fetch the complete content index at: https://www.thedaringcreatives.com/llms.txt > Use this file to discover other available public pages before exploring further. # Leah Fasten Trained an AI Model on 20 Years of Her Own Studio Portraits - URL: https://www.thedaringcreatives.com/creator-stories/leah-fasten-ai-photography-own-archive/ - Published: 2026-10-16T13:00:00.000Z - Updated: 2026-10-16T13:00:00.000Z - Description: Leah Fasten (@leahfasten) has shot commercial portraits for 20-plus years. Now she trains AI models on her own archive, casts people who don't exist the way she casts real ones, and posts every experiment, tools and prompts included. - Author: William Smith - Tags: Creator Stories, artist profile, ai photography, photography, adobe firefly, custom models, midjourney, nano banana, #medium:Still image, #method:Mixed, #handle:@leahfasten, #tool:Firefly, #tool:Midjourney, #tool:Nano Banana, #tool:Flux, #tool:HeyGen, #tool:Premiere, #issue-2026-10 In the mid-2000s, Leah Fasten put an Imacon V96C digital back on her Hasselblad 503CX. It was heavy, it topped out at ISO 400, and it glitched constantly. When she showed work made with it, people pushed back on whether it counted as photography. Twenty years later she trained an AI model on the glitchy files that back produced. Leah is a commercial photographer and creative director based in San Francisco and New York, and she posts as [leahfasten on Instagram](https://www.instagram.com/leahfasten/?ref=thedaringcreatives.com). Her photography clients, listed on [her photography studio site](https://www.leahfasten.studio/?ref=thedaringcreatives.com), include Bloomberg Businessweek, Fortune, IBM, MIT and Harvard Business Publishing. Since January 2026 she has been running [Okay, Unscripted!](https://okayunscripted.substack.com/p/okay-unscripted), a Substack she calls her GenAI sketchbook. That sketchbook is why she's worth studying. Almost every post names the tools and walks through the steps, and some print the prompt. You get to watch a working portrait photographer figure out, in public, which parts of 20 years of shooting carry over. ## Python bored her Her start with AI was rocky. She told the photographer Sarah Deragon, in [an interview on Sarah's Substack](https://sarahderagonphotographer.substack.com/cp/188520083), that she first tried Runway around 2021 for a flower series. Then she stopped. *"But making a data set, training models? It was so hard and expensive back then."* She came back in late 2022 with Stable Diffusion and Python, building training data out of her own photo archive. The images were fine. The work wasn't her kind of work. In [a post about casting](https://okayunscripted.substack.com/p/casting-as-existential) she wrote that it was *"way more about writing code than any sort of creative act. It bored me."* What hooked her came in early 2023, when the generations started looking like her photographs. She told Sarah: *"I realized: this is going to change photography the way film to digital did."* On [her AI portfolio site](https://www.leahsaiart.art/?ref=thedaringcreatives.com) she says her first real project was portraits of women who don't exist, a series she called "Real Women of AI." She also co-founded a generative AI community that grew to 3,000 people before she handed it off. ## Training Firefly on her own archive Leah is part of Adobe's GenAI VIP community, which gave her early access to Firefly Custom Models. She'd struggled with Firefly's base model because it didn't look like her work. So she trained [a custom model on her studio portraits](https://okayunscripted.substack.com/p/i-used-adobe-custom-models-in-firefly), using about 30 images from studio shoots with people of different ages, ethnicities and styles. She built versions trained only on student portraits and versions that mixed students with professionals, and she liked the mix better. The results in that post are straight out of the generator, with no retouching. She played with what the T-shirts say. ![A collage of studio portraits of students against colored backdrops, wearing tees that read GENERATIVE COLLEGE, UNIVERSITY and ARTIFICIAL INTELLIGENCE](https://storage.ghost.io/c/4b/dc/4bdc4be6-5abc-4e80-986d-59016a3f1831/content/images/2026/10/leah-fasten-firefly-custom-model-portraits.jpg) Leah Fasten — unretouched generations from a Firefly custom model trained on her own studio portraits, via Okay, Unscripted! Then she went back to that old digital back. In [her post on the V96C](https://okayunscripted.substack.com/p/glitching-in-the-mid-2000s) she trained a Firefly model on glitchy images the back made, glitches included. She draws a straight line between the pushback she got for going digital early and the way people talk about generative AI now. *"I didn't care. I loved that digital back, glitches and all."* ![A peach rose in a sunlit garden, sliced by bands of pink digital glitch](https://storage.ghost.io/c/4b/dc/4bdc4be6-5abc-4e80-986d-59016a3f1831/content/images/2026/10/leah-fasten-v96c-glitch-model.jpg) Leah Fasten — generated with a Firefly custom model trained on glitchy files from her Imacon V96C digital back, via Okay, Unscripted! She uses the same move on herself. For a May 2026 post she [generated a younger version of herself](https://okayunscripted.substack.com/p/prompting-with-love-not-fear) with a custom model in Firefly, then prompted Nano Banana to put that kid on roller skates at a plaza in Massachusetts in the 1980s. The model added the Star Market sign on its own. That's the image at the top of this page. ## She runs it like a shoot The thing I like most about Leah's posts is how much of the vocabulary comes straight from a production. She casts talent and lights the set, and she calls the extra frames outtakes. For castings, she generates talent in Midjourney, with style training done through Midjourney moodboards that mix her own photographs, generated images and light tests. Then she makes casting Polaroids from the selects with Nano Banana inside Firefly Boards. *"I've been casting real people for 20 years. This is the same process—moodboards, style references, light tests—just a different tool."* Lighting is harder. In [her first lighting test](https://okayunscripted.substack.com/p/genai-photo-studio-light-tests) she admits she still doesn't know many of the conventional lighting terms, because her lighting has always been improvised on set. So she prompts the way she'd talk to an assistant: *"the big northfacing window light covers her whole face."* Her editorial portrait series uses [phrases lifted from job descriptions](https://okayunscripted.substack.com/p/i-used-phrases-from-job-descriptions), and she printed the full prompt. In Midjourney: > "an award winning retro magazine editorial portrait of a hip person at work who {add your phrase from the job description here}" Then she takes those generations into Firefly Boards as image references, with a second prompt asking for an outtake of a full-page magazine portrait about that person's job. She also keeps the accidents. On [a lip oil image](https://okayunscripted.substack.com/p/generative-play-im-creating-with), the first generation came back with a weird white strobe across the model's face. She went with it. *"There is a particular kind of deadness that shows up when I try to force a specific outcome for a generative system."* ![A laughing woman with curly hair on a swing, backlit by bright sun with rainbow lens flare across her face](https://storage.ghost.io/c/4b/dc/4bdc4be6-5abc-4e80-986d-59016a3f1831/content/images/2026/10/leah-fasten-human-natural-light.jpg) Leah Fasten — made in Midjourney and finished in Firefly Boards with Nano Banana, from her post "Human Is the New Natural Light," via Okay, Unscripted! That swing image went with [an essay on the word "human"](https://okayunscripted.substack.com/p/human-is-the-new-natural-light). The prompt called for strobes, Portra film and editorial outtake energy. Her point in the essay: clients have asked for "natural light" for years while pulling her strobe-lit images as reference. She thinks "human" in a brief is starting to work the same way, describing a feeling more than a process. ## The journal practice Some of her work isn't trying to look like photography at all. For years she's done a version of morning pages and freewriting. Now she feeds fragments of that journal into Midjourney, two or three times a week, she told Sarah. The style training for [the journal images](https://okayunscripted.substack.com/p/visual-journal-february-2026) mixes her archive, her illustrations and earlier journal generations. She writes that these entries barely count as photography, and that it feels freeing. *"That space between language and image is I think where truth actually is."* That idea about space runs through [a February post on why GenAI images feel fake](https://okayunscripted.substack.com/p/authenticity-in-the-age-of-generative). As a photographer, she says, what you record is the distance between you and the subject, and that's where trust gets built. *"If that space isn't handled with intention and care, I believe it's impossible to build trust with a viewer."* ## What GenAI can do that a set can't One question she keeps asking is what problems these tools can solve for real productions. [Her post on kids](https://okayunscripted.substack.com/p/genai-images-of-kids) is the clearest example. Her custom model has no kids in it, only adult portraits and some academic settings to set the style. She generated six kids in Firefly with it, then used Nano Banana inside Firefly Boards to pull them into one group shot. After that she generated outtakes of the kids working on the paper-flower backdrop and making faces. ![Six laughing kids posed in front of a wall of paper flowers](https://storage.ghost.io/c/4b/dc/4bdc4be6-5abc-4e80-986d-59016a3f1831/content/images/2026/10/leah-fasten-genai-kids-group.jpg) Leah Fasten — six kids generated with a Firefly custom model, composited into one group shot with Nano Banana, via Okay, Unscripted! Her reasoning comes from years of commercial shoots: child modeling *"can be disruptive, exploitative, and unfair, with exposure too early to a very grown up world."* She doesn't present it as settled. She talked it through with Manuel Sainsily, a friend from the Adobe community, and ended the post by asking readers what they think. ## Directing, with love Leah's other recurring theme is direction. In [a May post](https://okayunscripted.substack.com/p/prompting-with-love-not-fear) she describes her best photo jobs as the ones where a creative director explains the feeling they're after and trusts her to get there. She borrows Brené Brown's "Power With" versus "Power Over" to describe the difference. *"With GenAI, we are now all creative directors directing a generative AI creative process."* Her read on slop follows from that. It happens, she writes, when the person making it isn't directing in a "Power With" way, and tries to force one specific output instead of leaving room for surprises. Tools Leah credits herself - [**Adobe Firefly**](https://firefly.adobe.com/?ref=thedaringcreatives.com) — Custom Models trained on her own photos, illustrations and journal images, plus Firefly Boards for compositing - **Midjourney** — casting, places, journal images; style training through moodboards - [**Nano Banana**](https://gemini.google/overview/image-generation/?ref=thedaringcreatives.com) — compositing and outtakes, usually inside Firefly Boards - [**Flux**](https://bfl.ai/?ref=thedaringcreatives.com) — lifestyle and fashion composites - [**HeyGen**](https://www.heygen.com/?ref=thedaringcreatives.com) — the avatar that introduces her AI portfolio - **Premiere** — cutting the avatar video together She discloses her setup for the writing too: the footer on her Substack says the words are hers, with SEO help from AI. ## Still shooting The camera work hasn't stopped, and the two sides keep crossing. In July 2025 she posted [a self-portrait shot in her studio](https://www.instagram.com/p/DMapvPuRh9F/?ref=thedaringcreatives.com), wearing an outfit she sewed, and composited onto a background she made in Midjourney with a moodboard trained on her own photographs of the ocean. In April 2026 she launched [leahsaiart.art](https://www.leahsaiart.art/?ref=thedaringcreatives.com) as a home for the generative work, with [an avatar to introduce it](https://okayunscripted.substack.com/p/leahsaiartart-and-a-genai-avatar): a Firefly custom model of herself, images made with Nano Banana on sets from her Midjourney journal, the avatar built in HeyGen, and the whole thing edited in Premiere. In late May she announced a talk at Adobe Create Now on June 3, during San Francisco Design Week, alongside Sarah Deragon, on creative bravery, custom models and style training. The images for [her announcement post](https://www.instagram.com/p/DY3ELqOj%5Ff2/?ref=thedaringcreatives.com) came from Firefly custom models trained on her illustrations, photographs and Midjourney journal images. For the event's stickers, she [trained a model on her journal generations](https://okayunscripted.substack.com/p/im-speaking-at-adobe-create-now-in) and made a series of California landscapes with it. Then in September she set up five Godox strobe heads among the plants on her balcony and made herself a new bio photo, which she describes as handmade. She called the post [Proof of Life](https://www.instagram.com/p/Ddp9awjFNQU/?ref=thedaringcreatives.com). *"And my work has always been about what it means to be human. Real. Alive. It still is."* If you want more photographers working this seam, we've profiled [Caspar Jade, a Berlin magazine photographer](https://www.thedaringcreatives.com/creator-stories/caspar-jade-fictional-portraits/) who posted a fictional person every day for 15 months, and [Vagabond Diary's Charles Lopez](https://www.thedaringcreatives.com/creator-stories/vagabond-diary-photographer-ai-art/), who shot for GQ before turning to AI. Leah's experiments land on [Okay, Unscripted!](https://okayunscripted.substack.com/) as she makes them.