Ideogram
Fan one ad concept into a 30-asset paid-social matrix on Martini using Ideogram V3 — every aspect ratio, every CTA variant, every headline localization, with the copy rendered legibly inside the image. Ideogram is the only AI image model that handles in-image text reliably; for performance-marketing work where the headline IS the asset, no other model in the stack comes close. Pair it with a saved brand prompt, fan into ratio-specific nodes, and ship a Meta + TikTok + YouTube test matrix without a Figma round-trip.
Ideogram renders quoted text near-letter-for-letter. Write: ad with the headline "TRAIN HARDER, REST EASIER" in bold sans-serif on the upper third of the frame. The quotes are load-bearing — Ideogram treats quoted strings as literal typography directives. For CTA buttons, use the same pattern: "SHOP NOW" or "TRY FREE" in a contained button shape.
Save the brand voice in a Text node: "Lifestyle photography style, soft natural light, warm sunset palette, athletic apparel context." Wire it into four Ideogram V3 nodes — 1:1 (Meta feed), 4:5 (Meta vertical), 9:16 (TikTok/Reels), 16:9 (YouTube preroll). Each node concatenates the brand prefix with the per-asset headline prompt. Brand voice survives across the matrix; only the headline and CTA change per node.
Ideogram listens to typography directives. Write "bold sans-serif, oversized, top-third of frame, white on dark overlay." Skip generic "with text" — Ideogram defaults to small, off-center, mid-tone text. Specify weight (bold, light, condensed), size (oversized, body, fine print), placement (top third, lower left, centered), and contrast (white on dark, black on cream).
For A/B testing, duplicate the Ideogram node and change ONLY the quoted CTA: "SHOP NOW" / "TRY FREE" / "GET YOURS" / "JOIN TODAY". Keep the brand prefix and the visual subject identical. Ideogram returns near-identical compositions with only the button text swapped — exactly what performance marketing needs for clean A/B reads.
For multilingual campaigns, duplicate the canvas node per language and swap the quoted headline: "TRAIN HARDER" → "ENTRENA MÁS DURO" → "鍛えて休む". Ideogram handles non-Latin scripts in V3, including Chinese, Japanese, Spanish, French, German. The visual is identical; only the typography swaps. This is the cheapest path to localized paid social.
For animated ad units, generate the static Ideogram frame, then chain into a video model (Seedance 2 or Kling 3) with "subtle camera push, headline text remains static" — the video model preserves Ideogram's rendered text while adding 4-8 seconds of motion. Drop the result into the Martini sequence builder and export to Premiere/DaVinci as a native ad unit.
Hero static for Meta feed. Quoted headline + quoted CTA = Ideogram's strongest pattern. Subtle dark gradient prevents headline blowout on bright backgrounds.
Lifestyle photography of a runner at sunset on a coastal trail, warm golden palette, athletic apparel context, the headline "TRAIN HARDER, REST EASIER" in bold oversized sans-serif on the upper third of the frame, white text on subtle dark gradient overlay, CTA button "SHOP NOW" in the lower right, 1:1 aspect
Vertical cutdown. Same headline, repositioned for 9:16 thumb-stopping. The brand prefix carries lighting language across formats.
[brand prefix] + Same scene, vertical composition for TikTok, the headline "TRAIN HARDER, REST EASIER" centered top, CTA "SHOP NOW" bottom center, 9:16 aspect
CTA variant for A/B testing. Only the quoted button text changes; everything else identical for clean variant read.
[brand prefix] + Same scene, headline "TRAIN HARDER, REST EASIER" but CTA changed to "TRY 14 DAYS FREE", 4:5 Meta aspect
Spanish localization. Ideogram V3 handles non-Latin and accented Latin scripts cleanly. Fan into all target locales from one canvas.
[brand prefix] + Same scene, localized headline "ENTRENA MÁS DURO, DESCANSA MEJOR" oversized top-third, CTA "COMPRAR AHORA" lower right, 1:1 aspect
Quote every piece of in-image text. Unquoted text gets paraphrased; quoted text gets rendered near-letter-for-letter.
Specify font weight, size, placement, and contrast in words. "Bold oversized sans-serif on upper third, white on dark gradient" beats "with bold text".
For A/B test matrices, change ONLY the quoted CTA between variants. Keep the brand prefix and visual subject identical.
Ideogram V3 supports magic prompt expansion — set magic_prompt to "ON" for exploratory drafts, "OFF" for production work where you need exact prompt control.
Use the style_type parameter (REALISTIC for photo ads, DESIGN for graphic-heavy CTAs, RENDER 3D for product mockups). Pin per asset class.
For high-stakes campaigns, run 3-4 generations per variant — Ideogram's text rendering is strong but not deterministic. Pick the cleanest typography.
Ideogram V3 returns 1024-2048 wide outputs with rendered in-image text holding ~95% legibility on quoted strings. Generation time 15-30s per variant. The text-rendering capability is what no other Martini image model matches — for ad creative where the copy IS the asset, Ideogram is the production tool. Output drops onto the canvas; chain to video models for animated ads, sequence builder for ad cutdowns, or NLE export for native delivery to Meta Ads / TikTok Ads / Google Ads.
Connect Ideogram V3 with other AI models on Martini's infinite canvas. No GPU required — start free.
Get Started FreeBlack Forest Labs
Generate the visual layer of an ad-creative matrix on Martini using FLUX.2 — lifestyle hero photography, product context shots, and aspirational scenes that feed downstream Ideogram nodes for headline overlay. FLUX.2 is the prompt-fidelity workhorse: it renders specific products, environments, and compositions almost literally, which is exactly what performance marketers need when the creative brief reads like a shot list. For text overlays, chain to Ideogram; for the photography itself, FLUX.2 is the cleaner pick.
View guideOpenAI
Generate ad creatives on Martini using GPT Image 2 — strong reasoning over compositional briefs and multi-element prompts, with reasonable in-image text rendering for short headlines and CTAs. GPT Image 2 sits between Ideogram (typography champ) and FLUX.2 (photographic prompt fidelity): it understands narrative briefs better than either, which makes it the right pick for ad concepts that require interpretation ("show a busy parent finding 5 minutes for themselves") rather than literal staging. Pair it with a downstream Ideogram node for the final headline pass when typography needs to be perfect.
View guide