Class 01.
Image generation.

What we will learn.

1. What GPT Image 2 can do, and where it still fails
2. Prompt structures from one word up to JSON
3. Twenty-plus prompt techniques across text, image, and multi-image work
4. Six categories of text-driven deliverables in a single prompt
5. Business use cases tied to real revenue
6. Layered workflows for professional brand systems
7. Higgsfield to access many models in one place
8. A hands-on assignment to build your own artifacts

GPT Image 2
against Nano Banana Pro.

The two leading image models in 2026. We will lean on GPT Image 2 as the primary, Nano Banana Pro as the comparison.

Token-based pricing.

Image input

$8.00 per 1M tokens

Cached image input

$2.00 per 1M tokens

Image output

$30.00 per 1M tokens

Text input

$5.00 per 1M tokens

Every reference image you upload is billed at the high-fidelity input rate, regardless of your output quality setting. Edit-heavy workflows cost more than generation-only workflows. Pricing accurate as of May 8, 2026.

Mr. Grateful.

Meta. Adobe. OpenAI. Now in San Francisco building my own AI startup. Three years teaching with Mindvalley.

Content. Examples. Use cases.
Techniques. Workflow.

Then an assignment at the end to make your own artifacts.

New capabilities.

Examples with transparent prompts.

Business use cases.

Real revenue. Real production.

Prompting.

Progressive overload. From one word, up to JSON, into thirty-two techniques.

Prompt structures.

Five formal frameworks. From one word to full JSON.

Simple text to image.

One word. One concept. Useful baseline.

Sunset

S.A.S.

Subject + Action + Setting. The simplest template that produces good results.

A monk meditating at sunrise on a Himalayan ridge.

Detailed prompt template.

What you do not tell the AI is left up to the AI to decide.

Subject. Action. Setting. Aspect ratio. Camera. Lighting. Composition. Mood. Style. Text. Negative.

A weathered fisherman in a yellow rain slicker mending nets on a wooden dock at golden hour. Shot on a Canon DSLR with an 85mm lens, f/2.8, shallow depth of field, rule of thirds composition with the subject on the right third. Warm overcast lighting from the left. Documentary editorial style. No watermark, no text overlay, no extra fingers. 3:2 aspect ratio.

JSON prompting.

For organization and isolation of variables. Easier to inspect, easier to compare, easier to debug.

{
  "image_type": "realistic candid smartphone photo",
  "aspect_ratio": "3:4",
  "scene": "warm living room in the morning, with large windows for natural light",
  "subject": "adult wearing a black t-shirt, natural candid expression",
  "action": "they hold a clear fancy glass filled with grape juice all the way to the brim",
  "required_details": [
    "Their left hand clearly shows exactly four fingers",
    "a classic wall clock in the background clearly reads exactly 7:14",
    "the image must look casual, realistic, and shot on a smartphone"
  ],
  "avoid": [
    "extra fingers",
    "wrong number of fingers",
    "wrong clock time",
    "blurry clock",
    "partially filled wine glass",
    "studio lighting",
    "cartoon or glamour look"
  ]
}

Natural language.

Don't overthink it. Just ask for what you want.

Make me a poster for my upcoming meditation workshop. Make it feel like a Studio Ghibli movie poster.

Baseline creator scene.

The simplest form. Pure text in, image out. No reference, no editing.

A creator filming a podcast episode in a sunlit home studio at golden hour. 16:9

Tip. Get a clean text-to-image working before adding any references or modifiers.

Time-of-day shift.

Provide an existing image. Ask for a modified version. The model uses the input as a foundation.

Transform this photo into the same scene but at sunset with warmer tones and a longer shadow.

Reference. Any starting scene. Tip. Describe what you want changed, not the entire scene from scratch.

Borrowing the look.

Use a reference image purely for its visual style. Generate a new subject in that style.

Use the visual style of this reference image: muted color palette, soft natural lighting, film grain texture, shallow depth of field. Apply that style to a new scene of a meditation altar with candles and incense.

Reference. A film still or photograph. Tip. Say "use this image as a style reference" out loud to remove ambiguity.

Borrowing the subject.

Use the reference for its subject. Apply a new style around it.

Take the subject from this reference photo and re-render them in the style of a Studio Ghibli watercolor.

Reference. A portrait or product photo. Tip. State which slot the reference fills.

Sketch to render.

Convert a hand-drawn sketch into a polished rendered version.

Convert this hand-drawn sketch into a polished 3D rendered version with realistic lighting and material textures. Preserve the composition and proportions of the original sketch.

Reference. A hand-drawn sketch. Tip. The fastest path from concept to client-ready visual.

Animation to realism.

Convert an animated character into a photorealistic version.

Convert this animated character into a photorealistic version. Keep the pose, costume, and expression. Use a professional DSLR look with natural lighting.

Reference. An animated still. Tip. Specify "DSLR look with natural lighting" to avoid uncanny results.

Realism to painting.

Reimagine a photograph as a classical or contemporary painting.

Reimagine this photograph as an oil painting in the style of John Singer Sargent. Visible brushstrokes, warm light, classical composition.

Reference. A photograph. Tip. Naming a specific painter gives the model precise visual vocabulary.

Style only, new subject.

Use the reference's stylistic vocabulary while generating a completely different subject.

Generate a portrait of a contemplative monk in robes using the exact color palette, lighting, and framing of this reference image. Same symmetrical composition, same muted pastel colors, same shallow depth of field.

Reference. A film still, painting, or photograph. Tip. Color palette, lighting, framing, texture, composition are independent levers.

Product placement.

Insert a specific product into an existing scene while preserving everything else.

Add a copy of "The Sacred Act" book on the table next to the candle. Match the lighting and shadows of the existing scene.

Reference. The original scene. Tip. Tell the model to match lighting and shadows explicitly.

Clothing item change.

Change a single clothing item without altering anything else.

Change the shirt to a deep navy linen button-down. Keep everything else identical.

Reference. A photo of the subject. Tip. "Keep everything else identical" is the magic phrase.

Object removal.

Remove an unwanted object and convincingly fill the space.

Remove the laptop from the desk. Replace the area with the same wood texture as the rest of the surface.

Reference. The original image. Tip. Tell the model what should fill the space, not just what to remove.

Color change.

Change the color of a single object while preserving texture and form.

Change the color of the sweater from gray to deep forest green. Preserve the texture and folds.

Reference. The original image. Tip. Be specific about the color name. "Deep forest green" beats "green."

Scene expansion.

Expand an image beyond its original frame to reveal context outside the crop.

Extend this image outward in all directions. Show the full meditation hall surrounding the seated figure, including the back wall, side altars, and the doorway behind. Maintain the lighting and atmosphere of the original.

Reference. A tightly cropped image. Tip. Tell the model what's plausibly outside the frame.

New environment, same subject.

Keep the subject identical. Replace only the environment around them.

Keep this person exactly as they are. Replace only the background with a sunlit ancient stone temple courtyard at golden hour.

Reference. A subject photo. Tip. Faster than a reshoot. Useful for headshots, product shots, content reuse.

Repair and colorize.

Repair damaged photos, colorize black and white, remove blur, upscale resolution.

Restore this damaged photo. Repair the tears, remove the discoloration, and colorize it as if shot on Kodachrome film in the 1970s. Preserve the original composition and the people's features.

Reference. A damaged or low-resolution photo. Tip. Describe what the photo should look like, not just what's wrong.

Isolated cutouts.

Clean isolated subjects on solid backgrounds for use as design assets.

Generate an isolated illustration of a brass singing bowl on a pure white background. Studio cutout style, clean edges, no shadow, ready for use as a sticker or design asset.

Tip. Specify "white background, no shadow, isolated" explicitly. Some models add subtle gradients by default.

Editorial layout.

Magazine cover with masthead, feature lines, hero photo, full editorial design.

Create a magazine cover for "The Sacred Act Quarterly" issue 12, summer 2026. Hero photo of a meditation hall at sunrise. Feature lines: "The Quiet Revolution," "Practice in the Age of AI," "10 Teachers Worth Your Time." Include masthead, issue number, barcode, and price $12. Premium editorial design, large display serif headline, clean grid layout. 8.5x11 aspect ratio.

Tip. The more text fields you specify, the more the model has to work with.

Marketing collateral.

Movie-style poster with title treatment, tagline, and credits block.

Create a movie-style poster for an upcoming Mindvalley class titled "Render Their Dream." Hero image: a creator pointing at a glowing AI-generated landscape behind them. Tagline at top: "Turn one prompt into a paying customer." Credits block at bottom: "Starring Mr. Grateful, Directed by Curiosity, A Mindvalley Production." Bold display typography, cinematic lighting. 27x40 movie poster aspect ratio.

Tip. Specify the exact text in each zone (title, tagline, credits) rather than leaving it to the model.

Educational material.

Recipe-style infographic with hero photo, step panels, metadata strip.

Create a recipe-style infographic poster for a contemplative morning ritual. Show a large, inviting hero photo of a morning altar setup on one side. On the other side, show several smaller step-by-step image panels that illustrate the practice from waking to first sip of tea, each with short captions. Include the practice title at the top, basic timing info, and a tools section. 16:9

Tip. Specify a hero element, supporting panels, and a metadata strip for clear hierarchy.

UI and software mockup.

Believable app interface with realistic UI text, exact numbers, proper layout.

Create a mockup of a meditation tracking app's home screen on an iPhone 16 Pro frame. Top status bar showing 9:41 AM. App header reads "Sacred Practice." Today's session card showing "Morning Sit, 22 minutes, completed." Streak counter at "47 days." Three buttons below: "Start Session," "Library," "Reflect." Bottom tab bar with home, library, profile, and settings icons. Clean iOS 19 design language, off-white background, dark text, sage green accent color.

Tip. Specify exact text strings, exact numbers, exact button labels. Vague prompts produce generic screens.

Diagrammatic content.

Whiteboard scene with handwritten diagrams, equations, labeled flows.

Whiteboard scene with the Render Their Dream framework written out in handwritten marker. Three big arrows: SIGNAL then RENDER then DELIVER. Under each arrow, three smaller hand-drawn callouts explaining the step. Hand-drawn diagrams, slightly imperfect handwriting, blue and black markers, slight smudges. Mr. Grateful is standing to the right pointing at the SIGNAL section.

Reference. A photo of yourself for character consistency. Tip. Describe the marker, the imperfection, the layout structure.

Stylized text treatment.

Text rendered as a stylized environmental element. Neon, signage, t-shirt graphics, tattoos, graffiti.

A neon sign hanging in the front window of a quiet teahouse at night. The sign reads "Be Here Now" in flowing handwritten cursive script, glowing soft pink against the dark interior visible through the window. Slight reflection on the wet sidewalk in front. Cinematic, moody. 16:9 aspect ratio.

Tip. Specify exact text in quotes, the script style, the color, the surrounding environment. Context sells it.

Different angles.

Same person, multiple angles, for use as a character reference sheet.

Take this reference photo of me and create three additional angles of the same person in the same outfit: a three-quarter view from the right, a profile from the left, and a back view. Same lighting, same background. Photorealistic.

Reference. Your portrait. Tip. Use a clean reference with neutral lighting and plain background.

Different environments.

Same person, multiple environments, identity stable across all of them.

Place this person in three different environments: a quiet meditation hall, a sunlit kitchen, and a misty forest path. Same outfit, same expression. Photorealistic. 16:9 aspect ratio for each.

Reference. Your portrait. Tip. State the lighting explicitly for each environment to keep the look coherent.

Pose and skeleton control.

Force a character into a specific pose using a separate pose reference.

Take the person in this reference photo and re-render them in the exact pose shown in this second reference. Maintain the original face, hair, and outfit.

Reference. Your portrait, plus a pose reference from posemy.art. Tip. Use a 3D mannequin pose, not another person's photo.

Age progression.

Same person at different ages, core features preserved.

Take this person and show three versions of them: at age 8, at age 35, and at age 75. Keep the same facial structure, eye color, and bone structure. Photorealistic.

Reference. Your portrait. Tip. Works best on clean front-facing portraits. Side angles confuse the model.

Animation frame generation.

Sequential frames of the same character mid-motion. Storyboard, comic, or animation.

Take this character and generate five sequential frames of them transitioning from sitting to standing. Each frame captures a different moment in the motion. Same outfit, same setting, consistent lighting.

Reference. A character or person. Tip. Use 3 to 8 frames. Longer sequences accumulate visual drift.

Multi-asset composite.

Combine separate isolated assets (people, objects, environments) into one cohesive scene. The crown jewel.

Combine all these reference assets into one cohesive scene set in the meditation hall reference. Show the woman wearing the prayer shawl while seated on the meditation cushion. Show the man holding the singing bowl while burning the sage bundle. The mala beads are draped on the altar near the candles. Use the meditation hall reference image without changing the angle or perspective.

Reference. A grid sheet of isolated assets plus an environment reference. Tip. Build model sheets first, then composite.

Reverse engineer to JSON.

Deconstruct any reference image into a complete JSON prompt that could regenerate it.

Analyze this reference image and produce a complete JSON prompt that would let me regenerate it. Include image_type, aspect_ratio, scene, subject, action, required_details, style (lighting, camera, mood, realism), and avoid fields.

Reference. Any image. Tip. The output JSON becomes your starting template for similar work.

Color grading and filters.

Apply a precise color grade to an existing image while preserving its composition.

Take this image and apply a Wes Anderson color grade: muted pastels, slight desaturation, warm shadows, cool highlights, gentle film grain. Preserve the composition.

Reference. Your starting image. Tip. Naming a director or film stock gives the model precise vocabulary. "Cinematic" is vague. "Kodak Portra 400" is not.

Upscaling.

Increase the resolution and detail of a low-quality image without inventing new content.

Upscale this low-resolution image to high resolution. Preserve all original details, sharpen the edges, restore fine textures. Do not invent new content or change the composition.

Reference. A low-resolution starting image. Tip. State "do not invent new content" explicitly.

Multi-step brand workflow.

Break a complex outcome into a sequence of focused steps. Each step's output becomes the next step's input. The bridge into Section 5.

Step 1: What is the professional logo design process for a wellness brand? Walk me through each phase.

Step 2: Apply that process to my company "The Sacred Act." First do the research phase: 200 words on visual conventions in the contemplative practice space, including what works and what's overdone.

Step 3: Based on that research, ideate 12 logo concepts for The Sacred Act. Number each with a short reasoning. Mix wordmarks, icon marks, and combinations.

Step 4: Take logo number [#] and create a color variation sheet showing it in the brand's primary palette plus three alternate moods.

Step 5: Take the final logo and create a mockup sheet showing it on a business card, a t-shirt, an app icon, and a website header.

Tip. Each step has a single focused job. The chain produces higher quality and gives you decision points along the way.

Workflow.

Create a brand using layers and professional process.

It is not about the individual capabilities.

It is about knowing how to layer the capabilities to get more sophisticated outcomes with control.

Research. Character creation. Branding. Ideation, decision, refinement, variations. For mood boards, colors, logos, characters, places. Anything really.

Six steps to a brand.

1. What is the professional logo design process? Imagine it says: research, ideation, decision, refinement, variation, mockup.
2. Do research on the psychology and design philosophy of logo design within my industry. And insight on how to stand out and have a logo that is cross-functional for social media and app icon.
3. Apply your research to ideate 12 wordmark and iconic mark logos for my company, numbered with your reasoning under each.
4. Upscale logo number [#].
5. Create a color variation sheet. Then a logo sheet with usage guidelines.
6. Create a sheet of mockups of this logo in various contexts.

Higgsfield.

All the models in one place.

Three reasons to use it.

Your turn.

Build your own artifacts before next class.