Table of Contents
Hey all, happy Fridizzle,
Quick James update since I know you're all deeply invested in the life of a one-year-old you've never met: the Prince for Kids show was a hit. The kid lost his absolute mind when they brought out the parachute. Full-body lunging for it, over and over, like it was the single greatest thing he'd ever witnessed. Which, to be fair, it probably was. He's one. The bar is low (or high… it’s all relative).
We also counted, and between daycare, family, and the party itself, this child consumed four separate cakes in five days. Four. He's been alive for one year, and he's already speed-running the American dream (prepping for Coney Island in July?). He also ripped his pants at the party, which I think we can all agree is the universal sign that things went well.
Anyway. We're heading up to Vermont this week, which currently has more snow than Salt Lake City and Colorado combined, so there is a very real chance the newsletter will be either late, short, or written from a chairlift. Consider this your advance warning. (Consider also that "written from a chairlift" is something I tell myself to feel productive while skiing, and that in practice I will simply not write it.)
I was planning to go deep on AI UGC this week, but then Kling 3.0 dropped, and I need to actually play with it before I say anything stupid. (This has never stopped me before, but I'm trying to grow as a person.) Sora 2 has been the move for AI-generated people, but Kling might change the math. So I'm parking that one.
Quick disclaimer: we don't actually use AI UGC in any of our client ads. But enough of you keep asking that I've gone pretty far down the rabbit hole.
Instead, let's talk about something I think more of you actually need right now: getting product insertion right in static ads. Getting the size and lighting right, and making sure your product doesn’t look like it was pasted in by someone who's never held a physical object.
I know it seems like a lot to have 6 steps, but it is worth it if you want to be consistent.
Let's get into it.
The 6-Step Product Insertion Workflow
I'm going to walk through each step with the tools, the prompts, and the logic behind why each one matters. If you just want the prompts, scroll down. I won't be offended. (I will be a little offended.)
Step 0: Create or Source Your Base Scene
You need an image to work with. Either without a product, or with a placeholder product, you'll swap out.
Your options:
Source from Pinterest, then recreate in Fal.ai. Find a vibe you like on Pinterest, then use an image-to-prompt tool (imageprompt.org gives you 5 free per day) to reverse-engineer the prompt, and recreate something similar in Fal. This avoids any copyright weirdness.
Generate from scratch in Nano Banana. Text-to-image. You describe the scene, it makes the scene. Pretty straightforward.
Use existing UGC or lifestyle photography. If you already have photos of people holding things, great. We can work with that.
The key here is that you want a scene that looks natural and has clear context for where the product should go. A hand holding something. A table with space. A person mid-sip. You get the idea.
From Pinterest
Reverse Engineered
Step 1: Remove the Existing Product (If Needed)
Tool: Nano Banana
If your base scene has a product in it that you want to replace, you need to remove it first. This is one prompt in Nano Banana:
Remove the [product] from [location/hand]. Keep the hand/scene natural and intact.For example: "Remove the can from his hand."
That's it. Nano Banana will remove the object and keep everything else looking natural. The hand stays. The grip stays. The product disappears.
With Can
Sans Can
Step 2: Manual Product Positioning in Canva or Preview
This is the step that makes everything work. This is the cheat sheet I was talking about. I know… labor… woof.
The process:
Upload your scene (product removed) into Canva
Upload your product image
Position the product in the scene at realistic scale
Use your own product knowledge to judge the sizing. You know how big your product is. You've held it. Trust that.
Export as a single flattened image
Why this works: You are giving the AI a scale reference it can match. You're removing the guesswork. The AI doesn't have to figure out how big a 12oz slim can is relative to a human hand because you've already shown it.
Err slightly smaller rather than larger. Oversized products always look fake. Always.
Canva Interface (Neat!)
No Can
With Can!
Step 3: Scene Analysis (Prompt 1)
Tool: Claude or ChatGPT
Now you upload that Canva composite to Claude or GPT and ask it to analyze the scene. You want it to break down the lighting, the shadows, the color palette, the whole vibe. This becomes the brief that informs the final composite.
I'll put the full prompt below, but the gist is: you're asking the AI to be a cinematographer. Tell me where the light is coming from. Tell me if the shadows are warm or cool. Tell me what the grain looks like. Tell me the mood.
I use this prompt a lot for more than just product insertion, so it’s probably a good one to have in the back pocket.
Analyze this image as a scene reference for product placement. Provide a comprehensive breakdown of:
Lighting:
- Primary light source direction, intensity, and color temperature (warm/cool/neutral)
- Secondary/fill light sources if present
- Quality of light (soft/diffused vs. hard/direct)
- Any practical lights visible in frame
Shadows:
- Shadow direction, density, and edge quality (sharp vs. soft falloff)
- Shadow color (pure black, warm, cool, or tinted)
- Ambient occlusion areas (where surfaces meet)
Color & Tone:
- Overall color palette and dominant hues
- Color grading style (lifted blacks, crushed shadows, cross-processed, etc.)
- Saturation level and any color casts
- Highlight and shadow tones
Aesthetic & Mood:
- Visual style (editorial, lifestyle, minimal, maximalist, nostalgic, etc.)
- Era or cultural references if applicable
- Texture quality (grain, noise, smoothness)
- Depth of field and focus characteristics
Environment Context:
- Setting type and implied time of day/season
- Surface materials and their reflective properties
- Spatial depth and perspective
Output this as a detailed scene brief that could guide realistic product compositing or AI generation of a matching product shot.
Output: Detailed scene analysis briefThis output becomes the foundation for everything that follows.
Step 4: Placement Logic + Compositing Specs (Prompt 2)
Tool: Claude or ChatGPT (same conversation)
Now you upload the clean product shot alongside the scene, paste in the analysis from Step 3, and ask the AI to figure out exactly how to composite the product into the scene.
This is where it determines things like what angle the product needs to be at to match the camera. Which fingers should be in front of the can? Where do the shadows fall? What reflections should appear on the surface?
Again, full prompt below. But you're essentially asking it to be a VFX supervisor.
You are receiving three inputs:
1. A reference scene image
2. A product image to be inserted
3. A scene analysis brief (below)
PRODUCT ANALYSIS
Analyze the uploaded product image and describe:
- Product type and physical form factor
- Dimensions and proportions (estimated)
- Surface material and finish (matte, glossy, metallic, semi-transparent, etc.)
- Current lighting on the product (where highlights fall, shadow side, any reflections)
- Viewing angle and perspective
- Key brand elements (colors, logo placement, text)
PLACEMENT LOGIC
Based on the scene, determine:
- The most natural insertion point (held in hand, resting on surface, replacing existing object, etc.)
- Required scale relative to scene elements
- Perspective/angle adjustments needed to match the scene's camera angle
- Occlusion mapping (what elements should appear in front of or behind the product)
- Hand/grip positioning if applicable
COMPOSITING REQUIREMENTS
Generate specific adjustments to make the product appear native to the scene:
- Light source matching (direction, intensity, color temperature)
- Shadow creation (cast shadow direction, density, softness, color)
- Highlight and reflection placement to match environment
- Color grading shifts to unify with scene palette
- Edge integration (contact shadows, ambient occlusion, feathering)
- Material-specific effects (condensation, environmental reflections, surface scattering)
SCENE ANALYSIS BRIEF:
[PASTE YOUR STEP 3 OUTPUT HERE]Step 5: Generate the Nano Banana Prompt (Prompt 3)
Tool: Claude or ChatGPT (same conversation)
Now you take all that detailed compositing analysis and compress it into a prompt that Nano Banana can actually use. Under 150 words. Direct imperative commands. No jargon.
The AI translates its own technical analysis into natural language instructions. "Match the warm overhead lighting from the front-left" instead of "adjust color temperature to 3200K with a 45-degree key light angle."
You are receiving a detailed compositing brief (below). Convert it into a concise, direct prompt for an AI image generation/compositing tool.
Output format rules:
- Write in direct imperative commands ("Replace," "Match," "Add")
- Combine related instructions into flowing sentences
- Prioritize the most critical information: placement, perspective, hand occlusion, lighting direction, color temperature
- Keep total prompt under 150 words
- Omit technical jargon the AI tool won't interpret (hex codes, percentages, pixel values)
- Translate specific numbers into natural language where possible (e.g., "slight," "subtle," "warm")
- Structure in this order: (1) Core action, (2) Perspective/scale, (3) Occlusion/grip, (4) Lighting, (5) Color/mood, (6) Finishing texture
Do NOT include:
- Explanations or rationale
- Multiple options or "if/then" conditions
- References to "the brief" or "as specified"
Compositing Brief:
[PASTE YOUR STEP 4 OUTPUT HERE]
Output a single continuous prompt ready to paste into Nano Banana.Step 6: Final Composite in Nano Banana
Tool: Nano Banana (image-to-image) on Fal.AI
You feed Nano Banana three things:
Your Canva composite (the scene with your manually positioned product)
Your clean product image
The prompt from Step 5
Set it to generate 2-4 variations.
And that's it. You'll get back product shots where the product is the right size, the lighting matches, the shadows make sense, and it doesn't look like someone discovered the paste tool in Photoshop for the first time.
Quick Tips Before You Go
On sizing: Trust your product knowledge in the Canva step. Use reference objects in the scene for scale (hands, faces, common items). And again: go slightly smaller rather than larger. Oversized products always look fake.
On hand-held products: Always specify finger occlusion explicitly in your prompts. Mention grip pressure and shadow contact. Natural grips are slightly imperfect, so don't aim for perfection.
On scene matching: Match the grain and noise. This is often what breaks the illusion more than anything else. If the scene has lifted blacks, lift the blacks on the product. Warm tones integrate better than cool ones nine times out of ten.
When it doesn't work: Regenerate with slight prompt variations. Adjust your Canva positioning and re-run. Try a different base scene angle. It's not magic. It's iteration.
This works great for: hand-held beverages, food products, beauty and skincare in-use shots, tabletop placement, and lifestyle UGC-style content.
That's It For This Week
If you got the Wumbo reference in the thumbnail, congratulations, you were raised right. If you didn't, it's from SpongeBob, and I refuse to explain further. (Fine. It's the episode where Patrick convinces SpongeBob "wumbo" is a real word and uses a belt to shrink all of Bikini Bottom. It's basically a documentary about AI product sizing.)
If you have questions, feel free to message me.
Have a great weekend!
Will