13:41
AI image upscaling
Magnific AI is an upscaling platform that uses generative AI to hallucinate photorealistic fine detail — skin texture, hair strands, fabric weave — onto soft or low-detail images, but the amount of invented detail is governed by a "creativity" slider that, pushed high, starts changing the subject's identity and inventing content that was never in the source image, which the video argues stops being "photography."
Magnific AI adds generative-AI detail (skin texture, eyelashes, embroidery, hair strands, pores) to images that previously had none, demonstrated first on a soft Midjourney-generated portrait.
The video runs a real-vs-AI image guessing test using an Adobe Stock photo and an AI-generated image to show how soft AI output normally looks compared to an overly retouched real photo.
Magnific settings include an optional guiding text prompt, "Optimized For" presets (portraits, films and photography, video game assets, 3D render), and a choice of engine (Automatic vs Sharpen).
After Magnific upscaling, the presenter brings the image into Photoshop and applies Retouch4me plugins — Heal for blemishes and Dodge & Burn for automatic high-end shading via a soft-light blend layer — to tone down the AI's over-detailed look back toward realism.
Tested on a real (non-AI) photo, Magnific adds convincing new texture but changes who the person is — "it is not the same person."
The creativity slider (demonstrated at 2 vs 8 vs 10) is the main control: low values refine existing details faithfully, high values let the AI hallucinate new objects, textures, and even alter facial structure, eye shape, and teeth.
At creativity 8 the AI can produce bizarre errors (oranges appearing inside eggs); at creativity 10 faces started appearing on inanimate objects (a bowl), severe enough that the presenter says he couldn't show it because it would be censored.
Images can be upscaled a second time via "reuse upscaled image," and settings can be copied across images via "reuse parameters"; a large upscale costs 15 credits, a small one costs 5.
Tested on a landscape photo: the AI invents new trees and even a farmhouse that wasn't in the original, prompting the presenter to ask whether the result still counts as photography.
Tested on the presenter's own face (a 1080p screenshot) and on a screenshot of GTA San Andreas's CJ (using the 3D render and films-and-photography presets) — in both cases the identity is altered from the source.
The presenter's stated biggest gripe is pricing: it's an expensive monthly subscription with a capped, non-purchasable credit allotment, forcing him to sign up with a different email once credits ran out; the plan shown costs $39.
Magnific AI upscaling — An AI upscaling platform that uses generative AI to synthesize new fine detail (skin texture, hair, fabric weave) into an image rather than simply interpolating existing pixels. Apply: Upload a low-detail image, pick an optimization preset and creativity level, and run the upscale to produce a version with generated fine detail.
Creativity slider — A 0–10 Magnific setting controlling how far the AI is allowed to deviate from the source image, from faithful detail refinement at low values to invented, hallucinated content at high values. Apply: Keep it low (e.g., 2) to preserve the original subject and add plausible texture; raise it (e.g., 8+) only when accepting that the AI will invent new objects or alter the subject.
Optimized For presets — Content-type presets in Magnific (Portraits, Films and Photography, Video Game Assets, 3D Render, etc.) that tune how the upscaler interprets and renders an image. Apply: Select the preset matching the image's actual content, e.g. "3D render" for a video-game character screenshot, to get a more appropriate style of added detail.
Prompt-guided upscaling — An optional text-prompt field in Magnific that steers the kind of detail the AI generates during upscaling. Apply: Describe the subject in the prompt (e.g., "Japanese woman wearing a red lipstick") to help the AI generate contextually consistent detail.
Engine selection (Automatic / Sharpen) — Alternate Magnific upscaling engines, with Automatic suited to most cases and Sharpen offering a different detail-rendering style. Apply: Default to Automatic; try Sharpen and compare results when a crisper rendering is wanted.
Reuse upscaled image (double upscaling) — A Magnific feature that lets an already-upscaled output be fed back into the tool for a second upscaling pass. Apply: Click the button on a finished upscale and choose "reuse upscaled image" to run it through upscaling again for additional detail.
Reuse parameters — A Magnific feature that copies the full settings (engine, creativity, preset, prompt) used on a prior upscale onto a new image. Apply: Select "reuse parameters" from a previous result to apply identical settings to a different image without reconfiguring the sliders manually.
Retouch4me Heal — An AI-driven Photoshop plugin that automatically removes blemishes from a portrait. Apply: Run it as an automatic first pass on a retouching layer whenever visible blemishes need removing.
Retouch4me Dodge & Burn — An AI Photoshop plugin that automatically performs high-end dodge-and-burn shading as an adjustable-strength soft-light layer. Apply: Duplicate the layer, run the plugin, dial the blend amount down from full strength (e.g., to ~16%), check "soft light layer," and apply to subtly restore natural shading after an over-detailed AI upscale.
The creativity slider doesn't just scale intensity on the same output — it changes what kind of operation is happening, from restoring plausible lost detail (low) to inventing new content outright (high), which is why the presenter starts questioning whether high-creativity results still qualify as photography.
The workflow treats detail-adding and detail-moderating as two separate necessary steps rather than one: Magnific overshoots into unnaturally hyper-detailed skin, so a second corrective pass in Photoshop (Retouch4me dodge/burn at a low blend percentage, soft-light layer) is needed to pull the result back toward believability.
Identity drift shows up as a systemic property of the tool across very different subjects — a stock photo, the presenter's own face, and a video-game character — not as an isolated glitch in one image.
Failure modes don't scale linearly with creativity: bizarre but survivable artifacts (oranges in eggs) appear around 8, while at 10 the hallucinations cross into content severe enough to require censorship, suggesting a threshold rather than a gradual slide.
The credit-based pricing model is framed as a deliberate business constraint (no option to simply buy more credits) rather than an incidental cost of GPU-heavy AI processing, strong enough that the presenter routed around it with a second email account.
«so that my friend is generative AI on steroids but when it takes too much steroids or you do too much this is what you get so we definitely need to strike a balance»
— 00:06
«the photo on the left is actually a real photo from Adobe stock quite overly retouched and the photo on the right is an AI generated image»
— 00:46
«no this is not sponsored at all the platform is very very expensive given that you have to do a couple of trial and Evers and you have limited credits»
— 01:30
«there's one problem with this it is not the same person and that is the thing with this AI it changes the person quite a bit»
— 05:41
«when the creativity is high it turns into Gordon ramzi on steroids»
— 08:43
«where the heck did those oranges come inside of the eggs»
— 08:50
«when I turned at 10 faces started to appear on on the bowl so much so that I couldn't show it in this video because it would be censored»
— 08:56
«this is absolutely not me»
— 11:07
«my biggest gripe with this is the pricing it is just insanely expensive»
— 12:28
Reception
Strong interest in the AI upscaling technology drives high engagement on early comments, but extensive complaints about pricing and perceived clickbait create a divided audience despite appreciation for the creator's honesty.
A hands-on demo that makes Magnific's generative detail synthesis look genuinely striking in before/after comparisons, but the video's own findings — identity drift on faces, hallucinated objects at high creativity, and a capped-credit paid model — position it as an early, imperfect, and costly tool rather than a finished retouching solution as of December 2023.
13:41