A.I. / LLM Good RENDERING Prompts - well maybe


para-CAD
 Share

Recommended Posts

21 hours ago, Rich_Winsor said:

A taste of what Invideo can do with a Chat GPT image.

Got this figured out. Here is a sample of the quality of videos

you can create from your ChatGPT created images. This one

was only 3.44 MB.

 

Original ChatGpt image created from Chief Architect CPU Ray Trace image:

421714208_ChatGPTImageApr30202612_15_18AM.thumb.png.497786f083bb2bf4f1b575da2be9ebe2.png

 

Invideo video created from ChatGPT image:

 

Edited by Rich_Winsor
  • Upvote 1
Link to comment
Share on other sites

 

Glenn Dawg shoots the curl

 

864519786_ShootingtheCurl-resize.thumb.png.dd6487c1d100c85039d1b4076a89e7f9.png

 

1) combine image 1 (beagle on surfboard - don't ever mention Snoopy)

    with image 2 (Laird Hamilton dropping into a 30 footer as Jaws) in PSP

    to create image 3.

2) Take image 3 into ChatGPT and create image 4.

3) Take ChatGPT image into Invideo AI to create the following video.
 

 

Interestingly enough when creating the video with Invideo AI

the image creation was stopped @ 93% because it found a

questionable image. The software restarted the video creation

and the video failed to generate for the second time @ 93%

 

I didn't know how to proceed so I typed "what's the problem?"

and entered the prompt. Chat gurgled, thought about it and

said it was creating 2 videos for me to compare, 1 from GROK

and 1 from KLING 3.0. I didn't care for the GROK version but

I really like the KLING version. So I guess what I have is an

image generated in KLING but sold to me thru Invideo. 

 

Hadn't heard about them. Here's their website.  https://kling3.io/

 

Lastly could somebody please look in on Mike G. Just make sure

he seems OK and check the area for any sharp objects. :ph34r:

 

 

                                                                                                                                                                                                                  

 

  • Like 1
  • Upvote 1
Link to comment
Share on other sites

A.I. / LLM Good RENDERING Prompts

 

Here you go, straight from the horse's mouth.

 

Question posed to ChatGPT:

"How to structure a prompt to get the best image?" 

 

(It turns out that the order in which you address

the elements is crucial to getting the best results.)

 

Response: (scroll all the way to the bottom for a quick, good overall prompt template)

 

A strong image prompt usually has 5 parts, in this order:

  1. Subject — what the image is about

  2. Environment — where it takes place

  3. Composition / Camera — framing, angle, lens, distance

  4. Style / Lighting — cinematic, photorealistic, painterly, etc.

  5. Constraints — what must NOT change or appear

A reliable formula looks like this:

[Subject] in [environment],
shot as [camera/composition],
with [lighting + mood + style],
showing [specific actions/details],
while maintaining [important constraints].


Example Structure

Basic Prompt

A mechanic working on a vintage truck inside a clean modern garage, wide-angle shot from eye level, warm cinematic lighting, photorealistic detail, realistic tools and grease marks, shallow depth of field.


Better Prompt = More Specific

Compare:

Weak

A garage with mechanics.

Strong

Two mechanics servicing a lifted black pickup truck inside a high-end automotive workshop, polished concrete floors, organized tool chests, hydraulic lift visible, photographed with a 35mm lens at eye level, realistic fluorescent overhead lighting mixed with warm sunlight from the open bay door, ultra-photorealistic textures, cinematic color grading.


The Most Important Sections

1. SUBJECT

Be extremely clear about:

  • Who or what

  • How many

  • Appearance

  • Action

Example

A young woman in a red evening dress opening double front doors and smiling warmly.


2. ENVIRONMENT

Describe:

  • Location

  • Time of day

  • Weather

  • Background objects

Example

Upscale suburban home at night with warm interior lighting and stone entryway.


3. CAMERA / COMPOSITION

This dramatically improves results.

Useful terms:

Camera Angle

  • eye level

  • low angle

  • overhead

  • close-up

  • medium shot

  • wide shot

Lens Style

  • 24mm = dramatic/wide

  • 35mm = cinematic natural

  • 50mm = realistic portrait

  • 85mm = compressed portrait look

Composition

  • centered composition

  • symmetrical framing

  • shallow depth of field

  • background blur

  • locked-off camera


Visual Examples of Camera Styles

Wide Cinematic

Image

Image

Image

Image

Image

Image

Image

Portrait / Close Detail

Image

Image

Image

Image

Image

Image

Image


4. LIGHTING & STYLE

Lighting is often the difference between “AI-looking” and believable.

Useful styles:

  • photorealistic

  • cinematic

  • ultra-detailed

  • documentary style

  • natural lighting

  • soft ambient light

  • moody shadows

  • golden hour

  • overcast realism

Example

Soft warm practical lighting with realistic shadow falloff and cinematic contrast.


5. CONSTRAINTS (VERY IMPORTANT)

This is where you control the AI.

Examples:

  • Do not alter image dimensions

  • Maintain original framing

  • No extra people

  • Keep camera static

  • Preserve architecture

  • No text or watermark

  • Keep proportions realistic

This prevents the model from “getting creative” in bad ways.


Best Professional Prompt Format

This format works extremely well for photorealistic results:

Create a photorealistic image of [subject] in [environment].
Camera is [composition/lens/angle].
Lighting is [lighting description].
Include [important details].
Maintain [constraints].
No [undesired elements].


Example Using Your Style of Projects

Create a photorealistic image of a luxury residential front entrance at night.
A young woman is opening one of the double front doors and smiling warmly toward the camera as if greeting a guest.
Camera remains locked-off and centered with symmetrical composition using a 35mm cinematic lens.
Warm interior lighting spills naturally onto the porch while exterior lighting remains soft and realistic.
Maintain the exact architecture and framing of the original image.
No additional people, no text, no exaggerated AI features, no camera distortion.


Advanced Trick: Priority Order

The AI pays more attention to:

  1. Early prompt text

  2. Repeated concepts

  3. Concrete details

  4. Visual language

So:

  • Put critical requirements FIRST

  • Put restrictions LAST

  • Repeat important concepts once


Excellent Keywords for Realism

These often improve quality:

  • photorealistic

  • cinematic

  • natural skin texture

  • realistic lighting

  • physically accurate

  • detailed textures

  • shallow depth of field

  • film still

  • realistic proportions

  • HDR photography

  • documentary realism


Things That Commonly Ruin Images

Avoid vague phrases like:

  • “make it cool”

  • “awesome”

  • “beautiful”

Avoid contradictory terms:

  • “dark bright lighting”

  • “cartoon photorealistic”

Avoid overload:

  • Too many subjects

  • Too many actions

  • Too many styles at once


One of the Best Overall Prompt Templates

Create a photorealistic cinematic image of [main subject] in [environment].
Shot with a [lens] from [camera angle].
Lighting is [lighting style].
The subject is [action/expression].
Include [important environmental details].
Maintain [composition/constraints].
No [undesired elements].

That structure alone will improve image quality dramatically.

Link to comment
Share on other sites

If you want to dig a little deeper look into stage-aware prompting.

 

Question posed to ChatGPT:

What is stage-aware prompting?

 

Stage-aware prompting is a technique where you build an image or video prompt in layers (stages) instead of trying to force everything into one giant instruction.

Rather than saying everything at once, you guide the model through:

  1. Foundation

  2. Composition

  3. Details

  4. Motion / emotion

  5. Refinement

It works especially well for:

  • photorealistic images

  • cinematic scenes

  • consistent characters

  • image-to-video generation

  • multi-step editing


Why It Works

AI models often fail when prompts contain:

  • too many priorities

  • conflicting instructions

  • action + style + camera + environment all competing equally

Stage-aware prompting fixes this by:

  • establishing the scene FIRST

  • then layering complexity

  • then refining only specific elements

This produces:

  • more stable compositions

  • fewer hallucinations

  • better realism

  • more consistent lighting and anatomy


Think of It Like Filmmaking

A director does not say:

“Make an emotional cinematic masterpiece with realistic lighting and movement and perfect composition and…”

Instead they build:

  1. Set

  2. Camera

  3. Lighting

  4. Actor blocking

  5. Performance

  6. Final polish

Stage-aware prompting does the same thing.


The 5 Core Stages


Stage 1 — Foundation

Establish ONLY:

  • subject

  • environment

  • overall realism/style

Do NOT overload this stage.

Example

Photorealistic upscale home entrance at night with symmetrical architecture and warm interior lighting.


Stage 2 — Composition & Camera

Now define:

  • framing

  • angle

  • lens

  • shot type

  • camera behavior

Example

Locked-off camera, eye-level medium-wide shot using a 35mm cinematic lens. Symmetrical composition maintained throughout.


Stage 3 — Subject Action

Now add:

  • motion

  • pose

  • expressions

  • interaction

Example

A young woman slowly opens the right-side front door, smiles warmly, and gestures invitingly toward the viewer.


Stage 4 — Secondary Detail Layer

Now refine:

  • textures

  • realism

  • lighting nuance

  • physical accuracy

Example

Realistic skin texture, natural hand motion, physically accurate door movement, subtle cloth motion, cinematic shadow falloff.


Stage 5 — Constraints & Protection

This stage is CRITICAL.

Prevent the AI from drifting.

Example

Do not change the architecture, camera angle, framing, or lighting design. No extra people, no text, no exaggerated facial features.


Example: Single Giant Prompt vs Stage-Aware

Weak (everything mixed together)

A cinematic photorealistic woman opening a door smiling with warm lighting realistic shadows realistic motion symmetrical composition luxury home…

This confuses priority.


Strong (stage-aware)

Stage 1

Photorealistic luxury home entrance at night.

Stage 2

Locked-off symmetrical medium-wide shot with a 35mm lens.

Stage 3

A young woman opens the front door and smiles warmly at the visitor.

Stage 4

Natural body movement, realistic lighting spill, subtle shadow transitions.

Stage 5

Maintain exact framing and architecture. No additional characters or camera movement.

The model now understands:

  • what matters first

  • what is secondary

  • what must never change


Stage-Aware Prompting for IMAGE GENERATION

Excellent structure:

Stage 1 — Scene

Modern luxury garage interior.

Stage 2 — Composition

Wide-angle eye-level shot.

Stage 3 — Main Action

Two mechanics work on a lifted black truck.

Stage 4 — Realism

Natural fluorescent lighting with detailed tools and realistic grease textures.

Stage 5 — Constraints

Maintain original room dimensions and wall spacing.


Stage-Aware Prompting for VIDEO

This becomes even more powerful.

Because video models struggle with:

  • motion consistency

  • camera stability

  • identity drift

  • object permanence


Video Prompt Staging Example

Stage 1 — Scene

Front porch of an upscale home at night.

Stage 2 — Camera

Camera remains completely static and centered.

Stage 3 — Initial State

Both front doors are closed.

Stage 4 — Action Sequence

The woman opens the door, smiles, waves, and gestures inside.

Stage 5 — Motion Quality

Realistic hand motion and accurate hinge physics.

Stage 6 — Constraints

No camera movement, no cuts, no additional people.


Advanced Stage-Aware Technique:

“Anchor → Motion → Protection”

This is one of the strongest modern structures.


1. Anchor

Lock the visual world.

Symmetrical luxury entryway at night with warm lighting.


2. Motion

Introduce only the animation.

Woman opens the door and waves naturally.


3. Protection

Prevent corruption.

Camera remains static. Architecture unchanged. No object warping.


Another Advanced Trick:

Hierarchical Priority Prompting

Use wording that signals importance.

High Priority

  • must

  • exact

  • maintain

  • locked-off

Medium Priority

  • should

  • preferably

Low Priority

  • optional

  • subtle

Example:

Camera must remain static.
Architecture must remain unchanged.
Lighting should remain warm and realistic.
Subtle ambient movement is acceptable.

This teaches the model priority.


Visual Style References That Pair Well with Stage-Aware Prompting

Cinematic Realism

Image

Image

Image

Image

Image

Image

Controlled Motion / Human Action

Image

Image

Image

Image

Image


The Biggest Benefit

Stage-aware prompting reduces:

  • melted hands

  • drifting faces

  • warped architecture

  • random extra objects

  • unstable motion

  • inconsistent lighting

because the AI receives information in a structured hierarchy instead of chaotic prose.


Best General Template

Stage 1 — Environment
Stage 2 — Camera
Stage 3 — Subject
Stage 4 — Action
Stage 5 — Realism
Stage 6 — Constraints

This is one of the best professional-level prompting methods for image and video generation today.

 

 

  • Upvote 1
Link to comment
Share on other sites

  • para-CAD changed the title to A.I. / LLM Good RENDERING Prompts - well maybe

Please sign in to comment

You will be able to leave a comment after signing in



Sign In Now
 Share