11:00
Photoshop Generative Fill
The video argues that Photoshop's new AI "partner models" — especially Google's Gemini 2.5 Nano Banana, alongside Flux Context Pro — let users make complex, multi-part photo edits with a single natural-language text prompt applied to a whole-image selection, replacing the laborious manual-selection workflow Generative Fill previously required, and across four live demos the presenter finds Nano Banana most often preserves the original scene most faithfully while Flux Context Pro sometimes wins on realism or overall quality.
New partner models (Flux Context Pro, Gemini 2.5 Nano Banana) are now accessible directly inside Photoshop's Generative Fill via the contextual task bar, alongside Adobe's own Firefly Image Model 3 and Image Model 1.
The core workflow shift: select the entire image (Cmd/Ctrl+A) instead of making precise selections, open Generative Fill, choose a partner model, and type one natural-language prompt describing every desired change at once.
The presenter frames this as being able to "speak to the AI like you would talk to a normal person" rather than performing a technical selection process.
Example 1: removing all distracting people from a photo except the central subject and filling in his booth details, done via one prompt with Nano Banana; Flux Context Pro is shown as a comparison but changed the framing.
Example 2: removing a detailed wire fence and adding clouds, contrasted directly against the traditional method of manually brushing a selection over every wire before using Firefly Image Model 3 — shown as far slower.
Example 3: adding a Volkswagen bus while simultaneously changing the time of day to sunset and the season to fall, all in a single prompt; Flux Context Pro's result is described as more photorealistic/HDR in this case.
Example 4 ("complete overhaul"): removing a car and telephone wires, adding clouds, converting a road to a grass park, and adding people walking by, all in one prompt; the presenter calls the Flux Context Pro result on this example "absolutely the winner."
Partner models (Nano Banana, Flux Context Pro) generate one variation per click, whereas Adobe's own Firefly Image Model 3 generates three variations per generation.
The model can be swapped at any time from the Properties panel, and the presenter recommends "always using multiple models to see which generation you like."
Nano Banana's first pass on the fourth example left some telephone wires behind, requiring a second generation to fully complete the request.
The video is framed as day 7 of an ongoing 10-day "AI photo editing series," with free downloadable sample images and PSD files offered via a linked sign-up.
Generative Fill (Photoshop) — Photoshop's AI-powered fill feature, accessed from the contextual task bar, that generates or replaces image content based on a selection and a text prompt. Apply: Make a selection (or select the whole image), click Generative Fill in the contextual task bar, then choose a model and enter a prompt before clicking generate.
Partner models — Third-party AI image models — Flux Context Pro and Gemini 2.5 Nano Banana — newly integrated into Photoshop's Generative Fill alongside Adobe's own Firefly models. Apply: In the Generative Fill panel or Properties panel, open the model dropdown and select a partner model instead of the default Adobe Firefly model.
Gemini 2.5 Nano Banana — Google's AI image model, offered as a Photoshop partner model, which in the video's examples most consistently reproduced the original scene, subject, and framing with minimal unwanted changes. Apply: Select Gemini 2.5 Nano Banana from the model picker, select the whole image, and enter a single natural-language prompt describing every desired edit, then click generate (one variation per click).
Flux Context Pro — A partner AI model in Photoshop's Generative Fill that, per the video, sometimes altered framing more than Nano Banana but occasionally produced more photorealistic or HDR-looking results, and won outright on the video's most complex example. Apply: Select Flux Context Pro from the same model dropdown and re-run generation with the identical prompt to compare its output against other models.
Firefly Image Model 3 — One of Adobe's own native Generative Fill models, distinguished by producing three variations per generation instead of one, and used in the video as the traditional-technique baseline for comparison. Apply: Use it after making a manual selection (e.g., brushing over an object like a fence) rather than selecting the whole image, then click generate to receive three candidate results.
Firefly Image Model 1 — Another native Adobe Generative Fill model listed alongside Image Model 3 in the model picker; mentioned by name but not directly demonstrated in the video's examples. Apply: Select it from the same partner-model dropdown as an alternative option to Image Model 3 when using Generative Fill.
Whole-image select + single natural-language prompt — The video's central workflow technique: selecting the entire image and describing every intended edit in one conversational prompt, instead of manually selecting each individual area to be changed. Apply: Press Cmd/Ctrl+A to select all, open Generative Fill, pick a partner model, type a complete multi-part instruction (e.g., "remove the fence and add clouds to the sky"), and click generate.
The presenter positions the real innovation as an interaction-model change (whole-image selection + conversational prompting) rather than purely an image-quality improvement — the pitch is about removing the technical selection step, not just better pixels.
Cross-model comparison on an identical prompt is presented as a standard practice to adopt, not just a one-off demo technique, since results and fidelity to the original scene visibly differ by model and by example.
The four examples escalate in ambition (single removal → detailed removal → addition plus two attribute changes → four simultaneous edits) which structurally builds the case that prompt complexity is no longer a practical constraint.
Despite general praise for Nano Banana across the first three examples, the presenter's own comparisons are not uniformly one-sided: Flux Context Pro is preferred outright in the final, most complex example, and Nano Banana's output was imperfect (leftover wires) on that same example.
Viewer replies (outside the video content) note the demonstrated per-generation credit cost is not disclosed on-screen, and PHLEARN's own reply confirms credit-cost details exist only on the tutorial's webpage, not in the video itself.
A commenter flags a real-estate liability concern (removing wires/poles/streets on a house photo could constitute misrepresentation if used for a property listing) — PHLEARN's reply confirms the demo was for creative purposes only and agrees removing permanent fixtures would be inappropriate for real estate/commercial use.
«These new partner models allow you to access different AI engines directly inside of Photoshop giving you creative control and allowing you to speak to the AI like you would talk to a normal person.»
— 00:11
«Nano Banana did a much much better job.»
— 02:34
«One simple text prompt and it does all of our editing for us.»
— 02:45
«You would have to go over every single wire in this entire fence.»
— 03:20
«This is an incredible result from just one text prompt.»
— 05:04
«But you can see just how powerful these partner models are. You can literally tell them everything you want to do in one text prompt.»
— 07:02
«All in all, I got to say this is absolutely the winner.»
— 10:15
«Our brand new partner models are incredibly powerful and so easy to use. I really love that you can just talk to them like you would talk to a normal person, and I love you can just select your whole image. You don't have to go through a technical process of selecting the changes you want to make.»
— 10:29
Reception
Mixed-to-skeptical reception with strong criticism around cost transparency and ethical concerns overshadowing appreciation for clear instruction.
A practical, live-demo tutorial that credibly shows a real workflow shift (whole-image selection plus conversational prompting) rather than just hype, and it presents unedited side-by-side model comparisons that include cases where the newly-hyped Nano Banana model is not the winner; it functions as promotional content for both PHLEARN's course funnel and Adobe's partner-model rollout, and per viewer feedback it does not disclose on-screen that each partner-model generation consumes paid credits regardless of result quality.
11:00