The month I went from three commissions to eleven is the month my “process” collapsed. Until then, my studio was a DM inbox, a downloads folder, and my memory. Client asks for a character piece, I generate, we go back and forth in the thread, files fly around named final_v3_ACTUALfinal.png, and somehow money arrives. At three clients that’s charming. At eleven it’s malpractice.
If you’re taking money for generative work, you’re running a studio whether you admit it or not. Studios have pipelines. Here’s the one I built after the collapse, stage by stage, including the boring parts that saved me and the client-handling rules I learned the expensive way.
Stage 1: Intake that filters before you generate
Every bad commission I’ve taken went bad at intake. The client “wants something cool with their character,” you interpret, they hate it, and now you’re three revisions deep into a brief that never existed.
My intake form now asks for: subject and references (minimum three images, no exceptions), style targets (with links to pieces of mine they want it to feel like), aspect ratio and final resolution, intended use (personal, streaming overlay, commercial print), deadline, and budget tier. Two magic side effects: clients who can’t answer these questions reveal themselves before I’ve spent an hour of GPU time, and “intended use” starts the licensing conversation on day one instead of after delivery, when it becomes an argument.
Stage 2: The queue is a sheet, not a feeling
Each accepted commission becomes a row: client, brief link, style/model plan, price, deposit status, current stage, revision count, due date. Stages are fixed: Intake, Deposit, Exploration, Client Pick, Refinement, Revisions, Final Delivery, Paid.
I keep this queue in a grid because a grid is the only format that survives contact with a busy month. For a while that grid was a normal spreadsheet, and it worked until it didn’t: no statuses anyone could trust, and me re-sorting the same tab every morning. I eventually moved the queue onto this work management platform because it kept the exact spreadsheet muscle memory but let me share a live view with clients. Viewer access costs nothing, so each client gets a filtered view of their row and stops asking “any update?” in the DMs, which was honestly worth the $19 seat by itself. But the tool matters less than the rule: the queue is written down, every commission is a row, and the row is the truth. When a client asks where their piece is, you should be reading, not remembering.
Stage 3: Exploration, and the model bookkeeping nobody does
Here’s the stage that separates hobbyist chaos from studio practice: recording how you made the thing.
For every commission I log the full recipe per candidate batch: base checkpoint and version, LoRA stack with weights, embeddings, prompt and negative prompt, sampler, steps, CFG, seed, and the ControlNet or reference pass if there was one. In ComfyUI this is nearly free since the workflow JSON embeds in the PNG, but I still copy the recipe into the commission row, because in three months, when the client comes back with “I loved the first one, can we do her brother in the same style,” the PNG is in some folder and the row is findable in five seconds.
The LoRA stack deserves its own bookkeeping. My style LoRAs get plain version tags (charstyle-v3, lineweight-v2), and the commission row records which versions shipped. When you retrain a style LoRA, old recipes silently stop reproducing, and “I can’t recreate my own commissioned piece” is a professional embarrassment you only need once.
Exploration output: 8 to 12 curated candidates. Not 60. Curation is the artist’s job; dumping a grid of raw gens on a client is asking them to art-direct, and they will, forever.
Stage 4: Client pick, then revisions with a fence around them
The client picks one direction from the curated set. From that point, revisions are refinements of the chosen piece, not new explorations. This distinction, stated in writing at intake, is the single most protective sentence in my process: “Includes two revision rounds on the selected direction; new directions after selection are quoted as new work.”
Generative work is uniquely vulnerable to revision creep because clients know iteration is “cheap.” A traditional painter gets natural sympathy about rework; when your client knows a re-roll takes a minute, every “one tiny thing” feels free to them. It isn’t. Inpainting the hands, fixing the jewelry, re-running the face detail pass, upscaling, color grading: a “tiny thing” is 40 minutes. The revision counter on the commission row makes the fence visible. Round 1, round 2, then a friendly note that further rounds are billed. Clients respect a counter they were told about; they resent one they discover.
Also learned the hard way: collect revision notes in one message per round, not as a live stream of reactions. “Send all changes as a single list” turns eleven pings into one actionable row update.
Stage 5: Delivery like a professional shop
Final delivery is a package, not an attachment: full-res master, web-res version, the license terms as a one-page PDF, and (for commercial tiers) a short provenance note describing the process. Clean filenames with the commission ID. Delivered from the row, marked Paid, archived.
The bigger creative studios have this packaging discipline down cold, and it’s worth stealing from them wholesale; outfits like Dreamworks Plus write about creative production and delivery workflows at the agency scale, and almost all of it maps down to a one-person generative studio. The client’s last impression is the handoff. A tidy package quietly justifies your rate; a Discord attachment quietly argues against it.
The numbers that actually run the studio
Once the queue is a sheet, you get the studio metrics for free, and two of them changed my pricing:
Hours per commission, honestly counted. Curation, revisions, and client communication are the real cost centers, not generation. When I logged them, my “quick” tier was paying below minimum wage. The tier got repriced.
Revisions per client. A repeat client who averages four rounds is not a good client at standard rates, no matter how friendly the DMs are. The counter told me who to raise prices on. My gut never would have.
Start smaller than this
You don’t need all of it this week. Do three things: write an intake form, put every active commission in one grid with a stage column, and add the two-revision sentence to your terms. That’s an evening of work. The rest of the pipeline grows from those roots the first time a month goes sideways, and in this scene, with commission demand still climbing, the sideways month is coming. Better to build the studio before it arrives.