Output & Delivery

The same set of decisions has to be expressed differently per model. Midjourney uses double-dash flags on the end of the line. Gemini, Flux and SDXL don't — they read the prompt as prose plus their own parameter fields. Some parameters have no equivalent at all across models. Switching the target rewrites the prompt into the right syntax and drops or translates the parameters that don't apply, instead of leaving Midjourney flags sitting in a Flux prompt where they'd be read as literal text.

The wording stays; the parameters change shape. Anything the target model can't do is dropped rather than left in as dead text.

Pick the target before you fine-tune the parameter deck. Tuning Midjourney flags and then switching to Flux throws most of that work away.

FLASH runs generation without leaving Witness Labs. You pick a resolution — 512px, 1K, 2K or 4K — and a layout mode — standard, storyboard, character sheet or multi-view. The prompt you're already holding is what gets sent; there's no separate FLASH prompt to write.

Lower resolutions come back fast and cheap and are honest about composition. Higher resolutions take longer, cost more, and are for the final version rather than for figuring out what you want.

Draft at 512px or 1K while you're still deciding. Only go to 2K or 4K once the composition is settled — high-resolution exploration is the fastest way to burn credits for nothing.

Standard is a single image. Storyboard lays out a sequence of shots from the same prompt, useful for narrative work. Character sheet gives you a reference sheet of one subject — front, side, expression variants — for reuse. Multi-view renders the same subject from several camera angles in a single grid.

The three grid modes only pay off when the subject is stable. If the prompt drifts between panels, you have a Logic Gate or lock problem, not a layout problem.

Pair character sheet and multi-view with locked style, subject and lighting segments. Without those locks you get several different characters instead of one character several times.

The high-resolution path is the same FLASH pipeline running at higher output fidelity. It's meant for the version you actually keep, not for exploration.

Longer render times and higher credit cost per image than standard FLASH. In return you get output that holds up when zoomed or printed instead of falling apart on inspection.

Lock the composition at standard resolution first. Send the same locked prompt to high-res once, not five times with edits.

A CMF spec sheet captures the material decisions inside a prompt — materials, colours, processes and finishes — plus a sustainability score and a manufacturing feasibility rating. It's the document you hand to someone who has to actually build or source the thing, not just render it.

A structured summary rather than an image. It reflects the prompt it was generated from, so a sloppy prompt gives a sloppy sheet.

Generate the spec sheet after the material decisions have settled, not during exploration. It's a record of what you chose, not a way of choosing.

When you generate images, the results collect in the Gallery — a grid of everything you've made this session and, for saved work, from your Vault. It's the visual counterpart to prompt history: instead of scrolling text, you scan thumbnails.

Each thumbnail is clickable. Selecting one opens it full-size in the Viewer, alongside the prompt that produced it.

Use the Gallery to compare variations side by side before you commit. Judging a re-roll against its siblings is far easier than judging one image alone.

Storyboard, character-sheet and multi-view renders arrive as a contact sheet — the frames grouped as a single set so you can read them as a sequence rather than loose images.

The set stays together in the Gallery and the Vault, so a seven-shot sequence or a four-view turnaround is preserved as a unit.

Generate sequences with a locked seed so the contact sheet reads as one coherent shoot, not four unrelated frames.

Clicking a thumbnail opens the Viewer — the full-resolution image with its prompt, parameters and target model attached. From here you can re-open the prompt in the workstation, re-inject it as a signature, or send it onward.

The prompt travels with the image, so nothing you liked is ever lost to a thumbnail you can't reconstruct.

When a render is close but not right, re-open its prompt from the Viewer and change one control — you keep everything that worked.

FLASH opens as a modal overlay from the header on every tab. Your current prompt is what gets sent — there is no separate FLASH prompt to write and no mode switch to make first. Close it and you're back exactly where you were in the workstation.

The workstation state is preserved while FLASH is open. Renders land in the Gallery underneath, so closing the modal reveals the new thumbnails.

Keep FLASH for iteration checks — 512px or 1K to see if the prompt is working. For the version you keep, use the render path from the mode you built the prompt in.

FLASH runs behind a small set of image models — fast general-purpose models for cheap iteration, and higher-fidelity models (including GPT-Image-2 on eligible tiers) for finals. The picker is inline in the modal; the prompt stays the same, only the engine changes.

Fast engines return in seconds and cost a fraction of the credits. High-fidelity engines cost more and take longer, but hold up at scale.

Iterate on the fastest engine that still tells the truth about your composition. Only switch engines when the prompt is settled — swapping engines mid-exploration confuses which change fixed what.

The Gallery has a filter rail so you can narrow to a mode, engine or session. A slideshow mode plays the current selection as a full-screen sequence. Any set of thumbnails can be exported as a contact-sheet image, or as a ZIP of the individual files.

Filters are non-destructive — nothing is deleted, just hidden. Slideshow and export operate on the current filtered set, so a filter narrows what you ship.

Filter down to one session's renders before running a slideshow or exporting a contact sheet. It's much easier to read a set that came from one prompt evolution than a mixed Gallery.