Adobe's November 11, 2026 MAX conference opened its annual Sneaks batch — experimental features that stress-test where generative AI meets Photoshop-grade control. The Verge's walkthrough highlights Project Sight Span (walkable photos), Project Screen Test (talk-to-your-character auditions), Project Sound Bender (cursor-mixed sound effects), and Project Rinse Repeat (copy-paste retouching). None ship today — Adobe has described Sneaks as early experiments that sometimes grow into shipping products — and they signal what Adobe may fold into Creative Cloud or Firefly after the Gemini and Drive integrations already announced this week.
TL;DR
| Sneak | What it does (per Adobe demo coverage) | Builder takeaway |
|---|---|---|
| Screen Test | Voice-chat with an AI character to set persona, then render scenario tests | Consistent character bible before generative video |
| Sight Span | Flat image → explorable 3D set → reframe export | Aggressive camera move without full 3D capture |
| Sound Bender | Prompted SFX on video + 2D sound canvas mixing | Faster temp audio than manual library hunts |
| Rinse Repeat | Clone one subject's retouch to others with native PS tools | Batch portrait and product cleanup |
Sneaks vs shipped MAX news this week
Adobe MAX 2026 already delivered partner news we covered separately:
- Google Gemini in Acrobat and Drive connectors
- Disney multi-year Firefly alliance
- MoviEdit multi-shot editing research
Sneaks sit one maturity stage earlier — R&D demos with Adobe-provided stills in press kits, not changelog entries. Treat them as capability signals, not API guarantees.
Project Screen Test: characters you interview
The Verge describes Screen Test as replacing pure text prompts with a FaceTime-like audition:
- Seed traits — anxious, awkward, etc.
- Pick or prompt a voice, or upload reference audio
- Talk with the character to refine backstory, fears, movement quirks
- Generate animated screen tests from scenario prompts
Adobe's goal is consistent personality across generative video — a pain point we tracked when synthetic presenters and ads multiplied without continuity rules.
For studios comparing stacks, this rhymes with character-centric video research but adds dialogic steering instead of single-shot prompts alone.
Open questions Sneaks do not answer:
- Export format into Premiere or Firefly Video
- Voice likeness rights when you upload samples
- Whether Screen Test uses third-party LLMs beyond Firefly — Adobe did not publish architecture docs in the coverage we read
Project Sight Span: photos you can walk through
A camera with a green aperture ring, evoking Sight Span reframes from explorable 3D photo scenes
Sight Span generates a navigable 3D environment from a single flat image — photo, painting, or illustration. You move the camera to see synthetic geometry outside the original frame, then export a high-resolution reframe.
The Verge notes tradeoffs:
- Navigation view looks soft or pixelated
- Final export restores detail at the chosen angle
- Tattoos and faces stay recognizable after aggressive moves
That is spiritually adjacent to spatial reframing on phones, but Adobe pushes larger camera translations — closer to previs than a mild crop.
Who cares:
- Marketing teams reframing hero product shots without reshoots
- Previz artists blocking scenes from concept art
- Accessibility reviewers asking whether synthetic backgrounds misrepresent real locations
Adobe research scientist Gautham Mysore told The Verge that intuitive interaction also gives experienced creatives finer control — Sight Span and Sound Bender are examples where pros steer rather than re-prompt.
Project Sound Bender: mixing SFX on a canvas
Piano keys curling into a green ribbon, evoking Sound Bender's cursor-driven mix canvas for generative SFX
Sound Bender uploads a video clip, generates SFX from a text prompt (example: bubbling sci-fi cauldron), auto-syncs audio, then exposes a square canvas:
- Corners hold different AI sound ingredients
- Dragging the cursor morphs the mix in real time
- Corners can be reassigned to new effects
This is generative Foley with a continuous control surface, not a static WAV download. For podcast and short-form builders, compare DeepL Voice tone control — different modality, same UX trend toward continuous sliders instead of prompt roulette.
Project Rinse Repeat: copy edits, not copy pixels
Rinse Repeat targets group portraits and product grids:
- Retouch one face manually (heal blemishes, even skin)
- Apply the same operation logic to other people with one action
- Add a shadow to one product → propagate shape-aware shadows to siblings
Crucially, Adobe says edits remain native Photoshop layers — not a flattened generative paste. That matters for agency approval chains and for teams already fighting AI slop in workplace writing who need reversible toolpaths.
A green brush fills a leaf-shaped gap in a picture tile, mirroring inpainting-style Sneaks that stay inside Photoshop tooling
Firefly strategy: toys vs displacement risk
The Verge closes with the tension Adobe rarely states in press releases: if Sneaks graduate, generative fill could replace some stock textures, character rigs, and manual cleanup — even when creators dislike the shift.
Adobe's public counter-narrative is control: Sneaks make expert tools more tactile, not only cheaper. Whether that holds depends on:
- Pricing of generative credits vs token inflation across labs
- Legal clarity on training data (ongoing in EU and US policy posts we track)
- Interoperability with non-Adobe video models teams already run locally
Creator workflow questions Adobe still owes answers to
Professional adopters will ask sharper questions than demo applause:
Project Screen Test
- Can exported character bibles become Firefly custom models with version control?
- Who owns voice likeness when employees upload reference audio from paid actors?
- Does conversation logging stay on device for NDA clients?
Project Sight Span
- Does export include depth maps for compositing in Nuke, or only flattened PNG reframes?
- How does Adobe label outputs that invent geometry behind subjects — IPTC metadata, Content Credentials, or nothing?
- What happens to journalistic ethics when news photos gain synthetic rear angles?
Project Sound Bender
- Are mixes royalty-free for broadcast, or governed by generative TOS caps?
- Can users import existing SFX libraries to anchor corners instead of pure generation?
Project Rinse Repeat
- When propagation misfires on diverse skin tones, does undo roll back all subjects or per-person layers?
- How does the tool behave on RAW vs Smart Object workflows — Sneaks demos used retouched examples.
Until Adobe posts technical FAQs, treat Sneaks as UX research validated on stage lighting, not production misery.
Competitive context at MAX week
Adobe is not the only vendor stretching 2D assets into 3D navigable space. Phone OEMs ship mild spatial reframe features; game engines import NeRF-like scenes from phone scans. Adobe's pitch is single-image convenience inside Photoshop-family ergonomics — valuable when art directors lack Unity staff.
On audio, Adobe competes with DAW automation and startups selling text-to-SFX APIs. Sound Bender's 2D canvas metaphor is the differentiator: it feels like grading, not prompting. That may map cleanly to Premiere panel UX if the Sneak graduates.
For character AI, Meta's Muse sets consumer expectations on glasses + voice. Adobe's Screen Test targets production character development, closer to animation pre-production than personal assistants — unless Adobe later ties it to Firefly Video the way MoviEdit attacks long-form story cuts.
What Sneaks are not
- Not in beta download as of November 11 reporting
- Not documented APIs for Firefly developers
- Not a replacement for MoviEdit benchmarks on long-form story cutting
- Not confirmed to use only Adobe-hosted models — wait for technical blogs if you need data residency answers
Practical watch list for creative technologists
- MAX keynote recording — Adobe usually drops demo videos with higher fidelity than stills
- Firefly release notes over the next quarters — Sneaks that graduate change names
- Terms of use updates if Screen Test imports voice likeness
- GPU requirements if Sight Span exports depth meshes, not just PNGs
- Comparison tests against open video+audio stacks for temp audio and character consistency
How this fits the 2026 Adobe arc
2026 MAX week stacks enterprise partnerships (Google, Disney) atop research Sneaks. Partners solve distribution; Sneaks probe interaction design for generative tools — the layer that determines whether pros adopt or sandbox AI.
Teams building custom creative agents should watch whether Adobe exposes Screen Test personas via API or keeps them inside Creative Cloud. Until then, agent harnesses in our DESIGN.md template guide remain the portable pattern for character sheets agents can read.
Enterprise procurement checklist for Sneaks watchers
If you are evaluating whether to budget for Creative Cloud upgrades tied to Sneaks marketing:
- Require Adobe legal to confirm training data scope for each Sneak before pilots on client assets.
- Map Firefly credit pricing for generative refills versus stock subscription spend you would retire.
- Ask whether Gemini connectors announced at MAX (our partner post) subsume some Sneak capabilities in Acrobat instead of Photoshop.
- Run side-by-side tests with open video tools for temp animatics — Sneaks are not guaranteed to beat specialized models on motion fidelity.
- Document exit ramps if Adobe deprecates a Sneak after marketing cycle — Sneaks historically vanish quietly when research pivots.
None of these steps require Adobe to ship betas; they protect producers from demo-driven purchase orders.
Related on explainx.ai
- Adobe and Google at MAX 2026
- Adobe–Disney Firefly Foundry alliance
- MoviEdit multi-shot video editing
- Zonder Labs AI slop taxonomy
- DeepL Voice real-time tone control
- HyperFlow open video+audio weights
Feature descriptions follow The Verge's November 11, 2026 Adobe MAX coverage and Adobe preview materials cited therein.