Render farms: Flux volume for options, Max for the hero still
Do not Max the whole batch.
Do not Max the whole batch. A render farm's option sheet is a stills argument: this HDRI, this material, this camera height. Volume belongs on the cheap Flux 2 row. The hero still — the plate a client would print, pause, and sign — is Flux 2 Max. Animating, or Maxing, every rejected option is how you pay the expensive meter to discover you preferred option four.
The farm already knows how to light a turntable. What it should not do is treat every lookdev still as a hero generate.
Options die at three credits
Generate the rack as images. One prompt family, one camera, one asset — change only the variable the client is actually picking: finish, time of day, lens height. Approve against the brief, not against a vibe word.
Flux 2 Pro is the catalog text-to-image row billed at 3 credits flat per call, professional / high-quality, requires_image false, aspects 16:9, 9:16, 1:1, 4:3, 3:4. Rank 29 on the text-to-image board in this catalog (Elo 1206, listed as FLUX.2 [pro]). Min and max credits are both 3. That is the budget you can afford to throw away while lookdev is still a fight.
It does not have its own /models page. The published Flux 2 family page is Flux 2 Max, where Pro sits as a variant. Use Pro to iterate chips. Do not quote 4K because a different Flux row prints it — max_output_resolution is unset on Pro.
Text-to-image is the door. The Flux roster is stills until you leave it. Do not start in video to "see the material in motion." Motion will not correct a wrong metal.
Max is the signed plate
Once one still is signed off, that file is the hero. Flux 2 Max is the catalog text-to-image row billed at 7 credits flat per call, features named text-to-image, ultra-high-quality, photorealistic, detailed. Rank 13 (Elo 1227, listed as FLUX.2 [max]). Same aspect list as Pro: 16:9, 9:16, 1:1, 4:3, 3:4. requires_image is false. Released after Pro. It is still one call, still an image, still not a turntable.
Seven credits is more than three. That is the point of not Maxing the rack. A farm that runs twenty lighting options on Max spent the hero meter twenty times and still has to pick. Run the twenty on Pro. Spend Max once, on the winner, when the still has to look like a photograph of the part rather than a sketch of one.
max_output_resolution is also unset on Max. Generate at the aspect you will actually deliver. Square "for flexibility" is how every later crop becomes a new composition and a new material.
Motion is a later meter
If the signed plate must move — a slow orbit, light sliding across a brushed face — that is image-to-video after the yes. It is not a reason to Max the rejected options "so they are ready to animate." Rejected options should never be ready to animate.
Lock the still before you spend on motion. Video meters seconds. Image rows meter a call. A lookdev fight belongs on the call. A 30-second identity pass of the signed still is a hold of the material you already picked. A 30-second pass of option seven is a long take of the wrong metal.
Do not typeset the SKU code, the RAL name, or "not for production" in the generate. Burn that string after with a text overlay. Then you can swap the next option without regenerating the geometry.
FAQ
Can I Max every option and pick later?
You can. You will pay 7 credits a call to discover a 3-credit fact. Volume is Pro. Max is the signed plate. That is the cost shape.
Why not generate the options as video so the client "feels" the material?
Because feeling is seconds, and seconds are the expensive meter. Clients pick chips. Give them chips. Animate the winner.
Is Flux 2 Max a video model if the farm needs a turntable?
No. Max is text-to-image. A turntable is a later I2V job from the locked still. Do not prompt Max as if it were a 30-second pass.
Does rank 13 mean I should skip Pro?
No. Rank is a reason to argue lookdev at three credits, then spend the photoreal row on the one still that will be graded, paused, or shipped. Rank is not a reason to Max the batch.