Reference library
Your brand, as model context.
Most "brand-aware AI" is wishful prompting. Macro wires moodboards, palettes, hero examples, and tone-of-voice straight into the model as conditioned context — so every generation looks like it came from the same campaign, because in a meaningful sense it did.
All capabilitiesEverything the model needs to know.
A Macro brand kit is a living document. Drop in colors, type pairings, hero examples, and a tone-of-voice paragraph. Every generation across every campaign in this workspace inherits it.
Confident but not arrogant. Warm. Specific. We make verifiable claims, never hype. Use 'your skin' not 'consumers'. Always include the active that does the work.
- "Wake your skin up."
- "Six weeks. Visible glow."
- "Revolutionary breakthrough!"
- "Game-changing skincare."
References aren't inspiration. They're conditioning.
Each reference is wired into the model with a specific contribution — color, lighting, composition, faces, mood, texture. Up to 32 per generation. The weight bars show the model's last attribution split for this hero shot.
Reference
ReferenceReferenceReference
ReferenceReferenceSame prompt. Different model in your head.
Without conditioning, large models drift toward the average of their training data. With Macro's reference library wired in, every generation is forced toward your brand's specific corner of that distribution.
Skincare hero — single bottle, soft shadow, premium minimalist.
- Generic 'AI skincare' palette — no brand colors
- Random product silhouette, doesn't match the SKU
- Light is too bright, brand mood is moodier
- Type space implies a different category
- Brand-locked Glow primary as the dominant hue
- Correct bottle silhouette pulled from the brand hero
- Clamshell light contour from the lighting reference
- Top-down composition that matches the carousel system
Every brand, in its own room.
Each workspace gets its own scoped library. References, brand kits, and tone docs from one client never leak into another's generations. Critical for agencies running 30+ brands at once.
Multimodal context, done right.
Macro doesn't just stuff references into the prompt. Each reference is encoded into image-conditioned context the model accepts as ground truth — supported on Nano Banana, FLUX.2, Veo 3, and the rest of the roster.
- 0132 refs · ~140msEncode referencesEach reference image is encoded into multi-resolution conditioning embeddings (CLIP / SigLIP / proprietary).
- 02weightedApply brand kit weightsPalette gets a high-weight color-conditioner. Type pairings get OCR-rendered exemplars. Tone doc is parsed for prompt-level constraints.
- 03Nano Banana 2Run guided generationModel conditions every denoising step against the encoded brand context. Drift is corrected at each step, not just at the end.
- 040.94 fidelitySnap to brand kitFinal output is color-snapped to the locked palette and run through a brand-fidelity classifier before showing up on the canvas.
Library essentials
Built for fifty-brand operators.
Per-client scope
Each workspace has its own library. No accidental brand drift between Halo and Hebbia, ever.
Versioned
Every brand kit and reference set is versioned. Roll back to last quarter's hero look in one click.
Shared with the team
Comment on a reference, propose a swap, mark a hero as 'approved' — same workflow, no extra tool.
Pinned palette
Lock a color and Macro refuses to drift, even on long generation chains.
"We had a brand book nobody read. Now it's in the library and every generation respects it. The drift is gone."
Daniel Okafor · Head of Design, Studio Crepuscular
A new medium needs a new canvas.
14-day Pro trial. No card. Every model included.