Section 3: AI Image Generation — Nano Banana Pro's Nine Use Cases, Grok vs Nano Banana, Photoshop's Five AI Tools, Ideogram Logos, Image Ads, and Upscaling

The workflow principle: 'create your image first and then use an AI video tool to animate that image into a video clip', scene by scene, because stills are cheaper and far easier to edit. Nano Banana Pro is his one-tool choice because its edits are surgical - 'unlike most traditional AI tools that completely recreate the image... Nano Banana keeps the image almost identical to the original.' Two access routes: Google Gemini (free, daily limits, watermark) and PromptEdit.com (pay per image, no watermark, aspect ratio, resolution, generation count, history; drop to plain Nano Banana to save money).

(1) Basic edits: 'Remove the phone and smartphone gimbal from the desk'; 'colorize this image'; 'Open up the eyes of the girl who was blinking.' (2) Environment change, simple to complex, subject fixed. (3) Style transfer, whole or partial: 'keep the man exactly, exactly the same, but change the style of everything else.' (4) Learn from a reference - a headshot style, an on-brand hex colour; renovation redesigns (bathroom, basement, backyard); a couple reimagined 1920s to the future. (5) Combine elements from multiple photos with explicit image numbers: 'create a new image where the man from image one is giving the man from image two the silver plaque'; product-in-scene: 'the perfume bottle from image one placed in the beach environment similar to image two... ground level'; outfit swaps; style match: 'change the style of image 1 so it matches the cartoon animation style of image 2... both the man and the background.' 'Be very clear with which elements from which image... and in what order.' (6) YouTube thumbnails - a swap produced a visible view spike. (7) Camera angles for video consistency: 'change the camera angle so it is a close up shot showing only the man's feet as he walks'; when the new angle reveals an unseen face, add 'Use the face from image 2 as a reference.' Mismatched tile or trouser colour will 'subconsciously pop in the viewer's mind.' (8) Character consistency: 'Use the lizard man from Image 1... in hiking clothes at the top of a mountain in Switzerland.' (9) Superimposed instructions: colored rectangles or a rough sketch on the source, referenced by colour in the prompt ('following along with the basic sketch provided'). AI clones are use case 8's extension - a full section later.

PromptEdit's update replaces 'img1/img2' with an @-mention: type @ and pick any uploaded reference by name. Elements is a saved library of recurring people, products and places, each of which can hold several images (four angles of a face) for better consistency across generations. Combined with Nano Banana's multi-image behaviour this is how a brand keeps one product and one presenter identical across a campaign.

Reasons to keep Grok: speed (10-20 s vs 35-60 s), cost ('two to four times cheaper'), and looser restrictions (PG-13 fight scenes Nano Banana may refuse). Head-to-head: portrait, product shot (a tie), interior, fantasy, food, family campfire, UGC skincare ad with text, thumbnail character replacement (Nano Banana wins clearly), full thumbnail from two references, AI clone from one reference of Conor. Verdict: Nano Banana Pro for realism, complex prompt adherence and editing; Grok shows 'AI sleekness' - too-perfect skin - and drops details (object placement, star trails), and 'slaps' a reference in where Nano Banana 'uniquely combines.'

Pricing: $20.99/month standalone, $54.99 for Creative Cloud. (1) Generative Expand - extend the canvas for a new aspect ratio, no prompt, three variations. (2) Generative Fill - lasso a region and type 'Add a plant' or 'add a pen sitting on the notebook.' (3) Generate Background - isolate the subject, short prompt for a new background. (4) Distraction Removal - lasso the unwanted element; it identifies and erases intelligently. (5) Generate Similar - iterate variations on a generated element by editing its saved prompt in the layers panel. Alternatives: Adobe Firefly (~$10/month, limited free credits), Photopea (free, web, ad-supported AI), Canva (free tier, paid AI).

Accessed on fal.ai (Explore > search 'ideogram v2' > latest). Two strategies. Detailed: write the full brief (Anthony Gallo Productions example) so reruns can change only the typography, only the colour, only the icon. Broad: prompt vaguely for inspiration, then 'reverse engineer prompting' - upload the result to ChatGPT and ask for a detailed Ideogram prompt that approximates it, which becomes your editable base. ChatGPT can draft the first Ideogram prompt too. Same approach for a YouTube channel banner.

Business case: the graphic designers he used to pay can be replaced or scaled, and offering ad creative on a $500-1,000/month retainer is a service. Principles: stop the scroll; solve a problem or desire; a no-brainer CTA. Workflow in Nano Banana Pro: reference 1 the product (course box), reference 2 a face; generate the scene; add headline and CTA text in a follow-up prompt. Competitor swipe: upload an ad whose style you like and ask for yours in that style. ChatGPT-assisted concepting (a roofing company before/after). UGC-style skincare ad with exact on-label text. Full video ads come later in the course.

Upscaling serves AI images, old family photos, and above all image-to-video source quality. fal.ai's Clarity Upscaler: upscale factor, creativity, resemblance, guidance scale, inference steps. Magnific AI (subscription) is his favourite - Upscaler plus Mystic generation, Relight and Style Transfer; settings include HDR and fractality. Website graphics: paste the landing-page copy into an already context-primed ChatGPT thread, ask for a graphic idea, then let ChatGPT write its own image prompt with no feedback - his cure for 'decision fatigue' - and refine by chat and inpainting (the bad-audio/good-audio waveform example).