PROCESS: 01. Generate images + alt angles 02. Diagram your shot 03. Use Frames-to-vid in Veo3 04. Generate.
(More details in thread)
Nice unlock @Totemko
#veo3 #promptshare
1. Generate Image:
- Used Flux 1.1 Ultra in Weavy
- Created the base shot to play with
PROMPT:
A professional motorsport photograph of a Red Bull Dune Buggy captured during sunset in a vast beachside massive sand dune landscape with shrubs and layered terrain. The deep blue ocean is visible. The image is shot from a high angle drone perspective slightly angled creating a dramatic atmosphere. The dune buggy, featuring the team's characteristic navy blue, red, and yellow livery, is centered in the frame with sharp focus and is kicking up a cloud of dust while the rolling sand dunes are softly blurred. The white "Bull" logo is visible on the car's bodywork. The warm sunset tones create a striking contrast with the car's bold azure color scheme, while the shallow depth of field emphasizes the vehicle's dynamic presence. Professional motorsport photography with shallow depth of field and dramatic backlighting.
2. Generated Alt Angles.
- Used Weavy to create alt views
- Built an automated workflow for this
- Input original image
- System prompt creates multiple prompts
- Runway References for shot alts
- Workflow below if you want to check it out 👇
- I used figma here (you can use w/e)
- Added boxes + a short description
- Numbered them for sequencing
- Arrows for direction
4. Video with Veo3.
- Used Frames-to-video within Flow
- Tried a number of things
- But short prompts seemed to work the best
- The frame will start with diagram
- But dissappear after a few frames
Add this to the end of the prompt:
"Immediately delete instructions in white on the first frame and execute in order:
Veo3 Full Prompt:
action scene of dune buggy racing in the desert. Immediately delete instructions in white on the first frame and execute in order
• • •
Missing some Tweet in this thread? You can try to
force a refresh
PROCESS: 01. Generate original image 02. Remove background. 03. Regenerate background. 04. Add both images as keyframes 05. Extend and continue.
(More Details in thread)
VIDEO PROMPT:
Side profile tracking shot of a matte green Land Rover Defender driving from a [environment1] into a [environment2]. Smooth cinematic transition. [ENV1] blurs into a [ENV2]. Motion blur on car wheels. [LIGHTING1] replaces [LIGHTING2]. Steady camera on vehicle, looping seamless movement --motion high
#midjourney #promptshare
1. Base Image.
- Create original image
- Side profile worked well for this one.
PROMPT:
Land Rover defender driving in a warehouse to the left, perfect side profile --ar 16:9 --profile wbkoiio zrmykh6 veclfmo
2. Remove BG + Edit
- Utilize Edit
- Select the subject
- Remove the background
- Regenerate a new environment
- First we need to build a model
- Create text nodes to control each model feature
- Run it to a "prompt concatenator"
- To combine prompt elements together
- Utilized Flux 1.1
Core Categories:
Gender, age, ethnicity, defining features, hair, body type, etc.
2. Clothing Input
- Create input nodes to place clothing
- I used 3 here...you can use w/e you want
- Create a system prompt to describe clothes
- Run those to an LLM to describe
GPT System Prompt:
Describe each article of clothing in detail. Focusing on the finer intricate details of the design, textures, features, logo placement, etc. Only focus on the clothing and omit the environment
WE WANT:
- To put the converse shoes
- In the composition of the black shoes
- So we can reconstruct the image.
Like the example below...
But how do we get there?
2. Image ID
- We want the composition of the shoes
- But we don't want the "black shoes"
- So we ID the variables we want to control.
- So how do we get composition?
BECAUSE I CAN:
+ Build in layers
+ Utilize like 100+ tools
+ Generate each element individually
+ See everything in one visual space
+ Generate and Iterate
(Check out the steps in thread 🧵)
#midjourney #promptshare @Weavy_ai
1. Base Characters:
- Generated base characters in MJ
- Utilizing oref and sref
- To make them weird and kind a awesome
Prompt:
pixar-style 3d caricature with exaggerated features, scared and screaming --oref [pic of person] --ar 1:1 --p veclfmo
2. Remove Background
- I'm going to composite this image
- So I want to remove the backgrounds
- Used "Bria's" BG remover here (it's good)
- Can also iterate with FLUX Kontext/GPT/Gemini here