11:27
Photoshop
Any 2D shape — an emoji, typed text, or even a photo portrait — can be turned into a realistic, dramatically lit 3D-looking render by building the shape/silhouette in Photoshop first and then feeding it into Adobe Firefly as a separate style reference and composition reference, giving the user full control over framing, environment, and material while Firefly handles the photorealistic rendering.
Build the shape in Photoshop rather than uploading directly to Firefly, because that preserves control over framing, styling, and environment.
Firefly's style and the shape's composition are set independently: first generate style options from a pure text prompt (no shape), pick one, and mark it 'use as style reference'.
Then upload the Photoshop-made shape via the 'composition' section as a separate composition reference, and raise the 'strength' slider to control how closely the output must match that shape.
For a simple flat shape like an emoji, direct upload of the shape as composition reference works fine.
For more complex shapes like text/letters, a flat 2D colored shape does not read as 3D to Firefly, so a fake 3D silhouette must be built manually in Photoshop first.
The fake-3D-silhouette trick: duplicate the text/shape layer, offset the copy, and recolor overlapping/adjacent letters or parts with visibly different colors (e.g., dark red vs. bright red vs. black) so Firefly's shape detection doesn't merge touching regions into one blob.
Use the Polygonal Lasso Tool to manually fill in gaps/connections between the offset and original letters to complete the pseudo-3D silhouette.
Background elements like a horizon/floor are created with the Rectangular Marquee tool plus a non-destructive Solid Color adjustment layer above the background layer.
If Firefly generates a texture (e.g., sand) that doesn't visually merge with the shape, painting a solid color block (e.g., yellow) over the relevant area in Photoshop before re-exporting helps Firefly apply the material correctly.
Firefly results can be regenerated repeatedly ('keep on clicking generate after generate') and any result can be upscaled for higher resolution, which also costs a credit.
Free versions of Photoshop still grant 25 free Firefly credits.
The same shape-into-Firefly pipeline was tested on an actual portrait photo to produce a 'stone statue' effect; it does not reliably preserve facial likeness, though it works well for non-human statues/objects.
Style Reference (Firefly) — A Firefly feature where a previously generated (or uploaded) image is marked, via its pencil icon and 'use as style reference,' to define the material, lighting, and mood of subsequent generations. Apply: Generate a few options from a pure text prompt describing material/environment/lighting (no shape involved yet), pick the best one, and click its pencil icon to set it as the style reference before adding a composition.
Composition Reference (Firefly) — A Firefly input, found in the composition section, that lets you upload a shape or silhouette image so the generated output follows that shape's structure. Apply: Export your Photoshop-built shape (emoji, text, or silhouette) as a PNG and upload it in the composition section so Firefly renders the chosen style onto that exact shape.
Strength slider — A control in Firefly's composition settings that determines how closely the final generated image must match the uploaded composition reference shape. Apply: Push strength all the way up when you need the output to stay faithful to the exact shape you built (e.g., readable text or a specific silhouette); lower it if some AI-driven deviation from the shape is acceptable.
Quick-and-dirty 3D silhouette technique — A manual Photoshop method for faking 3D depth information in flat 2D text/shapes so Firefly can interpret them as three-dimensional, since a flat 2D color shape alone generates a flat-looking result. Apply: Duplicate the shape/text layer, offset the copy behind the original, then recolor overlapping or adjacent letterforms with clearly different colors (e.g., dark red, bright red, black) so touching regions don't merge into a single blob before exporting as the composition reference.
Polygonal Lasso Tool gap-fill — Using the Polygonal Lasso Tool to manually select and fill connecting areas between the offset copy and original shape to complete the pseudo-3D silhouette. Apply: On a new layer above the offset shapes, click around the gap between the original and duplicated letter edges with the Polygonal Lasso Tool, then fill the selection (Alt/Option+Backspace) with a color matching the adjacent differentiated shape.
Non-destructive Solid Color adjustment layer — A Photoshop adjustment layer used to set a background/floor color without permanently altering pixel data, so the color can be changed at any time. Apply: With the background layer selected, click the adjustment layer icon, choose Solid Color, and set it (e.g., light gray) to create an editable floor/backdrop behind the marquee-selected horizon shape.
Upscale (Firefly) — A Firefly feature that increases the resolution of a chosen generated image, consuming an additional credit. Apply: After generating a result you like, click upscale on that image to get a higher-resolution version (at the cost of one more credit) before final use.
The core creative-control trick is decoupling style generation from shape generation: because Firefly lets you set 'style reference' and 'composition reference' from two different images, the user can text-prompt-explore materials/lighting freely without touching the actual shape, then bolt the two together at high strength.
Firefly's composition detection appears to treat touching same-colored regions as a single merged shape, which is why the manual color-differentiation step (dark red vs. bright red vs. black) is necessary purely as machine-readable shape metadata, not as a visual/aesthetic choice.
The 'paint a solid yellow block over the area' fix for the sand texture implies Firefly's material application to a composition reference can fail to fully saturate a shape unless the reference itself is filled solidly in that zone — a workaround for a specific rendering limitation rather than a general technique.
The video explicitly and directly reports the model's weak point: applying this shape-to-render pipeline to a real face degrades identity/likeness ('does it look like him I'm not very sure'), while the same pipeline is reported to work well when the goal is a generic human/animal statue rather than a specific likeness.
Naming Photoshop documents accurately at creation time (e.g., 'Smiley', 'name 3D') is called out as a small workflow optimization because Firefly's file browser will then surface the right file quickly during upload.
«you can turn any shape into 3D with Photoshop and Firefly super easily it's a child's play»
— 00:00
«it is Limitless and you completely control how the shape is going to be»
— 00:09
«we could have directly uploaded this to firly but that wouldn't allow us to control the Framing and The Styling and the environment»
— 00:26
«even if you're using a free version of Photoshop or a different kind of Photoshop you still get 25 credits to play with absolutely for free»
— 02:23
«let's increase the strength all the way up like Fat Joe so that it's exactly similar»
— 03:28
«we just have to keep the colors in a way that it different differentiates»
— 05:57
«does it look like him I'm not very sure and this is kind of where it slightly messes up»
— 10:17
«but if you wanted to create something like a human statue or an animal statue it's just a amazing»
— 10:23
Reception
Strong audience appreciation for the creator's teaching and technique, with legitimate concerns about Firefly's output quality and some philosophical objections to AI-generated content.
The video is a concrete, step-by-step workflow demonstration across three escalating examples (emoji, text, portrait) that separates Firefly's style and composition controls and supplies a manual Photoshop workaround for text shapes, while directly flagging that the same technique loses facial likeness on real portraits.
11:27