Lore
Own Product In Ai Scenes
1 тема · 167 концептов · 18 источников
Книги пока нет — соберите её в Telegram: /read own-product-in-ai-scenes
Темы
Концепты
- AI video object replacement pipeline (reference frame, then video pass)
- Rough composite as input to a Qwen image model
- Clean plate generation with Flux
- Distorting the inpaint mask to avoid shape bias in Wan
- Writing the video prompt from the reference image
- Structured prompt hierarchy (the seven-part formula)
- Prompt layers exist to remove model freedoms
- Dedicated typography block (typography as a priority anchor)
- Exact product identity statement
- Composition-first anchoring
- High-quality multi-angle studio reference images
- Nano Banana Pro for product typography
- Typography persistence in video via active image references (Kling 3.0 Omni)
- Component-level color naming (true color accuracy)
- Generative AI rendering with control nets, reference image and prompt
- Inner Reflections' Skyreels + VACE merged model
- Control-pass choice as a freedom dial (outline vs depth vs blend)
- Outline pass in Blender (Workbench + Freestyle)
- Depth pass in Blender (Z pass)
- Free AI preparation workflow (outline and depth from a viewport render)
- Single start-frame generation (cloud or local)
- Single start-frame vs multiple references for style guidance
- Image sequence to H.264 video conversion in ComfyUI
- Free AI renderer ComfyUI workflow: setup and single-reference run
- Shot-length limit and batch chaining in the AI renderer
- AI rendering trade-off: speed for control (previs use case)
- LLM-assisted prompt refinement and restyling
- Multi-reference mode with character sheets and an environment
- Character sheets with Z Image Turbo
- Iterative batching with overlapping start frames
- RunPod cloud templates for low-VRAM users
- Iterate by seed and prompt instead of simulation
- Nano Banana Pro in Photoshop via the Luminel plug-in
- Compact and non-verbal inputs instead of long descriptions
- Google Maps coordinates as a location prompt
- Garment reference image for outfit transfer
- Prompt chaining on one image for consistent frames
- Generate, then correct: alternating model output with Photoshop edits
- Text-only generation with no image input
- Generating a custom background to use as a reference image
- Pose-preserving scene change
- Prompt-based lens zoom
- Prompt-based upscaling with Nano Banana Pro
- Subject masking and background replacement
- Field Blur with pins for depth of field
- Cinematic blue hour night restyle
- Local Gaussian blur patch fix
- Masked curves adjustment for background depth
- Sketch-to-photo conversion
- Background photo as additional reference for campaign control
- Re-feeding the sketch to correct outfit drift
- Mask restoration of lost face detail
- Curves color correction for reference drift
- 2D lighting-setup sketch to image
- Curves, saturation and hue color grade
- Flattened-document input with a hidden reference layer
- One-image consistent ad campaign pipeline
- Consistency as the trust signal in ad campaigns
- Hero image as the campaign anchor
- Sourcing the hero image from Ideogram's explore page
- Custom GPT for ad-variation prompts
- Spiel Magic Edit: variations from one source image
- Product-in-hand image upload
- Turning image variations into talking UGC video ads
- Spiel UGC builder: Easy mode vs Custom mode
- Custom GPT for viral-optimized scripts
- Audio realism as the last step that makes AI UGC pass
- Directing the avatar with an action prompt
- 11 Labs voice changer on the extracted audio track
- 11 Labs subtle background-audio generation
- Blender to ComfyUI SDXL image-to-image render pipeline
- Flat, shadowless, high-contrast base render
- Redundant colour carry-through in an image-to-image pipeline
- Florence-2 detailed captioning
- LLM caption rewrite for render prompts (in-graph local LLM)
- SDXL Prompt Styler (preset styles instead of hand-typed phrases)
- SDXL input image preparation (resize, brightness/contrast, upscale then resize back)
- Juggernaut XL checkpoint and the image-to-image sampler graph
- Choosing the base model by add-on ecosystem (SDXL vs Flux vs SD 3.5)
- Depth and canny ControlNet maps from the input image
- Multi-ControlNet stack with per-net strength (CR Apply Multi-ControlNet)
- K sampler settings: CFG, steps and seed
- Denoise tuning for image-to-image
- LoRA (low rank adaptation) for targeted improvement in the SDXL render graph
- Add-on compatibility rule: LoRAs and ControlNets must match the checkpoint's base model
- Final upscale with an upscale model (ComfyUI)
- Preview nodes for intermediate results in the ComfyUI graph
- Installing the workflow via ComfyUI Manager
- AI video inside a traditional animation pipeline
- Asset-first approach
- Colour-coded storyboard shot triage
- Loose boards
- Batch-of-four critique and trait merge
- Visual formula: watercolor location plus golden-hour lighting
- Style negotiated with video-model limits
- Test assets in video before production
- Audit the references before rewriting the prompt
- Subtle character acting: the performance gap
- Texture-load rule for props
- Image-to-image asset editing
- Prop sheets only for recurring props
- Literal-wording check for effects sheets
- Blender gray previs as video input
- Five-block video prompt structure
- Front-loading the prompt
- Shared reusable Claude skill for prompts
- Generate-and-compare review loop
- Always include characters in vehicle references
- Zoom out (Rule one)
- Screenshot-as-camera-reference
- Prompting small interactive props for consistency
- Previs detail: a two-sided cost for performance shots
- Swapping a single reference can destabilize other properties
- Reference blocks: personality and quality, not visual description
- Including the 12 animation principles in video prompts
- IC-Light relighting (Stable Diffusion 1.5)
- Frequency separation for detail preservation (native ComfyUI)
- Mask-based high-frequency blend
- Gaussian blur radius tuning in frequency separation
- Depth and lineart background generation
- White-spot light-source mask (automatic lighting)
- Hand-drawn lighting mask via Preview Bridge
- Staged relight-then-restore-detail pipeline for product photography
- Limits of the IC-Light + frequency-separation workflow
- Loaded lighting-shape mask (Load Image Mask)
- Hand-drawn detail-preservation mask
- Where to add upscaling in the relight-and-restore workflow
- Multiple Segment Anything groups for complex subjects
- One-click prompt-swap routine for a new product
- Start/end frame anchoring for label-safe product animation
- Kling 3.0 product-ad pipeline: product shot to finished clip
- Frame difference drives the choreography
- Seamless loop from swapped start/end clips
- Motion masks mid-clip text warping
- Closing negative constraints (blocking improvisation)
- AI UGC three-level ladder (beginner, intermediate, expert)
- Pinterest reference-image creator build
- All-models platform with Sora 2 Pro, briefed like a real creator
- Scroll behaviour: why ads should look like native organic content
- Native format plus customer-language creative strategy
- Format-to-prompt pipeline (TikTok content search, content analyzer, Claude)
- Format library and winning-angle library
- MCP link between Max Fusion and Claude for building ads
- Full Amazon listing image set with Nano Banana Pro (no photographer or designer)
- Amazon listing image types (about 22) and the subset demonstrated
- Switch Gemini to 'Thinking with 3 Pro' to reach Nano Banana Pro
- Multi-angle, close-up phone photos of the physical product
- Plain-language prompting instead of engineered prompts
- Corrective prompting: name the flaw and ask for a retry in the same chat
- Competitor reference-graphic prompt for mobile-optimized graphics
- New-chat reset when the model hits a wall
- The human supplies the commercial judgment (which benefits to sell)
- Quality bar: would you accept it from a paid designer?
- Features graphic with labeled icons
- Targeted text edit by chat
- Momentum / piggyback prompting across graphic types
- Step-by-step instructions graphic
- Scale check against other objects in the scene
- Split placement from appearance: a stand-in sets geometry, the image model sets looks
- Garbage 3D product mannequin
- Nine-panel context sheet
- Compositor annotation layers (labelled references)
- Master prompt with reference-image roles (fixed scaffold)
- Switchable scene-description notes
- Multi-mannequin composition
- Rotate the 3D node to set the product's view angle
- Queue several runs and pick the best
Источники