16:15
Photoshop
Photoshop's Generative Fill (in Photoshop Beta) can execute a very wide range of practical and creative photo-editing tasks — merging photos, expanding scenes, removing/adding objects, changing clothing or backgrounds, and generating entire images from text — in seconds instead of hours, but the video's own closing framing is that "AI is a tool to help you, it's not a tool to replace you": professional-grade results still require manual selection discipline, color/light matching, and hand-painting on top of the AI output.
Generative Fill (Photoshop Beta, powered by Firefly) fills a selection with AI-generated content, optionally guided by a text prompt, and returns three variations per generation.
Selection technique drives output quality: leave a buffer ("a little bit of the meat") of original pixels around the edge of every selection so the AI has context to blend into.
When merging two photos, match lighting direction first (flip one horizontally if needed) before generative-filling the seam between them.
The video walks through roughly 20 distinct use cases: combining photos, expanding/uncropping images (including a square Instagram crop), replacing a subject, fixing a generated face, removing a person, removing a lens flare, changing clothing, iteratively expanding a scene across multiple passes, adding hair, removing bystanders and their reflections, changing background terrain, adding jewelry, removing braces/regenerating teeth, removing glasses, relocating a structure into water with a generated reflection, generating compositing shadows, replacing the sky via Select > Sky, generating a full image from a blank canvas, attempting a rainbow (a stated failure case), adding muscles/bodybuilder physique, removing power lines, and compositing a person into an existing photo.
Generative Fill is also reachable for free via Adobe Firefly's web app for users without Photoshop Beta.
Fine details are weak spots: the AI repeatedly failed to render nose studs/rings correctly, and rainbow generation consistently produced a cartoon look no matter how many retries were attempted.
The full-composite demo (inserting the presenter into a photo next to Taylor Swift) layers many small generative-fill passes on top of manual curves/Hue-Saturation color matching, hand-corrected shadows, and dodge-and-burn — presented as the level of skill that "separates you as a professional."
Generative Fill — Photoshop Beta's core AI feature that fills a selected area based on an optional text prompt, returning three variations per generation. Apply: Make a selection leaving a buffer of untouched pixels at its edge, optionally type a prompt into the contextual taskbar, click Generate, and pick or regenerate among the three results.
Photo bridging/combining — Using generative fill to blend two separate photos into one continuous scene. Apply: Match the lighting direction between the photos (flip one with Free Transform if needed), then use the rectangular Marquee tool with Alt/Shift to subtract/add regions and generative-fill the gap between them.
Image expansion / uncropping — Extending a photo's edges beyond its original frame using AI-generated content. Apply: Enlarge the canvas with the Crop tool, select all (Ctrl/Cmd+A), subtract the original image area with the Marquee tool (Alt/Option), then generative-fill the new blank border.
Square-crop expansion for social media — A specific application of image expansion to hit a target aspect ratio (e.g., Instagram square) without cropping into the subject. Apply: Press C, select the Square aspect ratio, Alt/Option-drag to expand the crop area, then select and generative-fill the newly exposed edges.
Adobe Firefly (web) — A free, browser-based version of generative fill for users without access to Photoshop Beta. Apply: Open Adobe Firefly's website, use its generative fill/background-replace tool, and type a prompt (e.g., "forest") to swap a background without installing Photoshop.
Subject/object replacement — Swapping one subject or object in a photo for a different one via generative fill. Apply: Select the subject (or load a saved selection from a channel) and type a replacement prompt (e.g., "chimpanzee") into generative fill, then generate.
Targeted local regeneration — Re-running generative fill on a smaller sub-area to correct artifacts left by a prior, broader generation. Apply: Select just the flawed detail (e.g., the face after a body swap) and regenerate with a refined prompt (e.g., "chimpanzee face").
Object/person removal — Deleting an unwanted person or object and letting the AI reconstruct the background behind it. Apply: Select the element to remove, leave the prompt blank, click Generate, and try multiple generations since details like hand placement may render incorrectly at first.
Distraction removal (e.g., lens flare) — Cleaning small unwanted visual artifacts such as lens flare from an image. Apply: Select the artifact with a buffer of surrounding pixels and generative-fill it blank so the AI reconstructs the underlying pattern.
Clothing replacement — Changing what a subject in a photo is wearing using generative fill. Apply: Select the garment and type a new garment description (e.g., "formal shirt") into generative fill, then generate.
Iterative/chained scene expansion — Repeatedly cropping and generative-filling to progressively extend a photo far beyond its original frame. Apply: After each successful fill, press C to expand the crop further, reselect and subtract the filled area, and generative-fill again, watching file size as it grows quickly.
Adding hair — Generating hair on a bald or thin-haired subject via generative fill. Apply: Select an area above/around the head with enough space, type a description (e.g., "flowing hair"), and generate.
Crowd/reflection removal — Clearing unwanted bystanders from a scene along with their reflections so no visible artifacts remain. Apply: Select the unwanted people plus their reflections (and any shadows they cast) and generative-fill the area blank.
Background element replacement — Swapping one background material or setting for another while preserving the subject's contact shadows. Apply: Select the full background including shadows and stray elements cast on the subject, type the replacement (e.g., "sand"), and generate — the AI can recreate matching shadows automatically.
Adding jewelry — Generating accessories such as necklaces, earrings, or nose jewelry on a subject. Apply: Select the target area (neck, ears, nose) and type the jewelry type (e.g., "pearl necklace"); note the video found nose studs/rings a weak spot, with the AI often producing "no stud" or an unrealistic ring.
Dental retouch (braces removal / teeth generation) — Using generative fill as a starting point combined with manual painting to remove braces or regenerate teeth. Apply: Select the mouth/teeth area, type a prompt (e.g., "teeth no braces"), generate, then hand-paint remaining artifacts using a sampled-brush technique credited to Aaron Nace.
Glasses removal — Applying the same selection-and-fill approach used for dental retouching to remove eyewear. Apply: Select the glasses area and generative-fill it to remove them, following the same technique used for braces/teeth.
Structure relocation with generated reflection — Moving an object such as a castle into water and having the AI generate a matching reflection. Apply: Select the destination area (e.g., a lake), leave a buffer, and generate — the video notes lake reflections rendered more convincingly than sea reflections.
Compositing shadow generation — Using generative fill to create a subject's cast shadow in a composite instead of hand-painting it. Apply: With the Lasso tool, select the area where the shadow should fall (accepting the AI may regenerate legs/feet too), generative-fill it blank, and regenerate multiple times until a convincing shadow appears.
Sky replacement (Select > Sky) — A dedicated AI selection feature (Select > Sky) combined with generative fill to swap out a scene's sky. Apply: Use Select > Sky to auto-select the sky, then generative-fill with a descriptive prompt (e.g., "dramatic") to replace it without manual color matching.
Text-to-image generation on a blank canvas — Using generative fill to create an entirely new image from a text prompt on an empty canvas. Apply: Select all on a blank canvas, write (or have ChatGPT write) a descriptive prompt, paste it into generative fill, and generate.
Rainbow generation (documented failure case) — An attempted use of generative fill to add a realistic rainbow, which the video reports consistently failed. Apply: Select the sky area and prompt for a rainbow, but expect a cartoonish result across repeated tries; the video recommends painting it manually instead.
Muscle/physique editing (bodybuilder look) — Reshaping a subject's body, including adding visible muscle mass, via generative fill. Apply: Select the torso/limb areas needing added muscle while leaving hands untouched, use a specific prompt like "male bodybuilder" (the video notes the plain prompt "bodybuilder" gave unusable results), then separately select and regenerate the legs to match.
Power line/cable removal — Clearing distracting overhead cables from street photography using generative fill, supplemented by the separate Remove tool for leftover trails. Apply: Select all visible cables and generative-fill blank; for small leftover artifacts, create a new layer and paint over them with Photoshop's Remove tool.
Full person compositing into a scene — Inserting a person into an existing photo and blending them in using generative fill together with manual color and lighting correction. Apply: Place the person on the target photo, build curves/Hue-Saturation adjustment layers to match color and lighting, generative-fill the edges/clothing/shadow areas selectively (redoing them individually as needed), then finish with dodging, burning, and a color-correction layer to fix remaining mismatches.
Matching lighting direction between two source photos (including flipping one) is treated as a precondition for a convincing AI-merged bridge — the AI doesn't compensate for physically inconsistent light on its own.
Selections need to include not just the target object but its shadow, reflection, and any trace it left on the ground (e.g., grass flattened by a body) — otherwise the AI leaves telltale "ghost" artifacts behind.
The AI's reliability is uneven by category: it produced convincing water reflections when relocating a castle and handled sky replacement well, but was named as a repeated failure point for rainbows and small jewelry details like nose studs/rings.
Iterative expansion (repeatedly crop + generative-fill the new border) can grow a single photo into an entirely new invented scene, but the video flags that the file size balloons quickly as a practical cost of this technique.
Generative Fill can generate images from scratch on a blank canvas, and the video shows offloading the prompt-writing itself to ChatGPT rather than authoring it manually.
Even where generative fill succeeds, professional-looking composites (the Taylor Swift example) still depend on traditional Photoshop skills layered on top — curves/Hue-Saturation matching, hand-corrected shadow direction and darkness, and dodge-and-burn — since AI results still needed touch-ups the video describes as necessary rather than optional.
For hard-to-paint corrections like braces removal, the video positions Generative Fill only as a faster starting point, not a replacement for the manual sample-and-paint brushing technique it credits to Aaron Nace.
«leave a little bit of the meat for the AI to grab upon there you go»
— 00:34
«we do need to do a couple of Trials till the hand is conveniently inside the pocket or outside the frame»
— 04:05
«generator film does make it easier as a starting point but you do need that painting technique to fill up those artifacts»
— 08:47
«the case of rainbow you know it is just best to do it manually»
— 12:02
«we do need to know color matching techniques and that is why whenever AI comes or whatever happens learning the manual techniques are essential»
— 13:39
«we cannot just leave everything to AI»
— 15:10
«this separates you as a professional»
— 14:25
«AI is a tool to help you it's not a tool to replace you»
— 15:36
Reception
Viewers appreciate the tutorial quality and creator's expertise, but many are frustrated with Generative Fill's technical limitations and inconsistent performance.
A breadth-first, hands-on catalog of Generative Fill's range across roughly 20 real editing scenarios, useful as a technique reference — but its own framing (and the failure cases it documents, like rainbows and nose jewelry) treats the AI as a fast first pass that still needs manual selection discipline and traditional retouching skill to finish.
16:15