Adobe opens Creative Cloud to more AI models - and puts you in the driver's seat
Adobe just pulled back the curtain at MAX 2025 with a clear message: use the model that fits the job. Firefly gets a major upgrade, third-party models are in, and custom brand models are on the way.
If you create visuals, edit video, or ship campaigns, this is about control. Less guessing, more choosing.
What's new
- Firefly Image Model 5 (public beta): More realistic lighting, texture, and anatomy - and it now generates natively at 4MP before any upscaling.
- Third-party model integrations: Use partner models right inside Creative Cloud apps instead of being boxed into one system.
- Firefly custom models (private beta, waitlist): Train a model on your brand assets by dragging and dropping - no prompt gymnastics required.
Why this matters for creatives
Different briefs need different engines. Product beauty shots, gritty documentary frames, multilingual VO, animated concepts - one model won't nail them all.
Adobe is making the tool choose you, not the other way around. That means better quality, faster iterations, and fewer workarounds.
Third-party models now inside Creative Cloud
Beyond Firefly, Adobe is bringing in select partners so you can pick the right specialist:
- Upscale in Photoshop: Topaz Bloom and Topaz Gigapixel for generative upscaling.
- Voiceover: Topaz Bloom and ElevenLabs Multilingual v2 for multilingual VO work.
- Additional options: Gemini 2.5 Flash Image (Nano Banana), Black Forest Labs' FLUX.1 Kontext, plus models from OpenAI, Runway, Luma AI, Moonvalley, Pika, and Ideogram.
Meta is notably absent from the current lineup. Adobe's approach favors a curated roster over a free-for-all.
Firefly Image Model 5: where it fits
Use v5 when you need realistic people, accurate lighting, and clean material detail. The 4MP native output gives you more headroom before you upscale or composite.
Pair it with Photoshop and Lightroom for edits that don't crumble under retouching. It's built for production, not just concept art.
Custom models for brand consistency
If you ship assets at scale, this is the most practical update. Firefly custom models (private beta) let you train on your existing brand library - colors, lighting, composition, illustration style - by dropping in your own examples.
The goal: on-brand assets without rewriting prompts for every request. Faster approvals, fewer manual tweaks.
Suggested workflows to try this week
- Concept to client-ready: Generate concepts with Firefly v5 → refine in Photoshop → upscale with Topaz Gigapixel → deliver.
- Global social variants: Create a master video cut → generate multilingual VO with ElevenLabs → swap captions and aspect ratios → publish.
- Systemized brand assets: Build a mood board of your brand's best work → apply for the Firefly custom model waitlist → train a style model → standardize campaign deliverables.
- Fast storyboard/animatics: Use a partner image model for stylized frames → sequence in Premiere or After Effects → iterate with client feedback same day.
What this means for your stack
Adobe is avoiding vendor lock-in. You get Firefly for commercially safe output, partner models for specialist tasks, and custom models for brand-true production - all inside the apps you already use.
Pick your model by outcome, not hype. That's the shift.
Links and next steps
If your team needs a quick playbook: map tasks to models, standardize prompts, and set a QA pass for brand consistency. Then automate the repeatable parts. The best ideas still come from you - this just clears the runway.
Your membership also unlocks:
