ComfyUI Word-Art Workflow Development
Budget: $30 – $250 USD
1. Project Overview - Comfy UI Workflow
I run a design brand that sells high-quality word-art (script + decorative elements) on Etsy and other platforms. Samples can be viewed at this URL: https://inyobus.com/samples
I need a qualified ComfyUI developer to build a workflow that reliably generates sellable word-art images from text phrases, probably using Flux Schnell (or another model that would work well) with good spelling and consistent style.
Environment & Constraints
2. OS / Hardware
• Windows 11
• GPU: NVIDIA RTX 4060 Ti 16gb
• Plenty of disk space
3. Existing ComfyUI install
• Main install folder:
C:\Art Geyser\0 ComfyUI Engine
• Uses a virtualenv at C:\Art Geyser\0 ComfyUI Engine\venv
• ComfyUI-Manager is already present and used.
4. Target Models & Node Packs
Base model. Through my research it seems that Flux Schnell is a good suggested model for text, unless you have another suggestion that:
o works on RTX 4060 Ti 16gb
o has good text rendering capabilities. (Most important)
o is appropriate for typographic / sign / logo style images.
If Flux Schnell truly cannot be made to work reliably on this GPU, propose a close alternative with similar performance and typography quality, and document the choice.
Node packs / custom nodes
Please install & configure only what is needed, but at minimum:
1. Any Flux-specific loader / node pack required to run Flux Schnell, or whatever alternate model is used.
If additional nodes are needed (ControlNet, IPAdapter, etc.), list them explicitly and verify they are compatible with the versions you choose.
5. Functional Requirements – What the Workflow Must Do
High-level goal
Given a phrase like:
“FAITH OVER FEAR”
or a phrase like “Real courage is doing the right thing when no one applauds you” with a subline like an author’s quote “Charlie Kirk” the workflow should generate a word-art image like the samples I provide.
• Correct spelling at least 75–90% of the time.
• A consistent “sellable” style — clean, bold, legible typography.
• Embellishments like flourishes, borders, watercolor backgrounds hearts, dots or anything other relevant artistic embellishment.
This is not about random art; it’s about building a reliable typography engine that creates word art. These images can be used in a variety of products such as mugs, t-shirts, pillows and so on.
Image specs
Default output (can be configurable in the workflow):
• Size: 1500 × 1500 px (or 1024×1024 if Flux strongly prefers it, but document that)
• Mode: RGB
• Background:
o Mode A (default for now): pure white (#FFFFFF)
o Mode B (added from the prompt): light neutral / farmhouse-style textured background (beige, wood, etc, floral style, or any other background style described in the prompt. (most of the time not used.)
• File format: PNG saved into a configurable folder, with filename including the first three words in the phrase slug and seed (e.g., faith_over_fear.png). When there will be multiple phrases that are the same we will let windows number them (1)(2) etc.
6. Model-Driven Text
Flux Schnell (or similar) directly creates the letterforms and style.
• You may use a text-accuracy LoRA (like Texta or equivalent) if compatible with Flux or your chosen model.
• The prompt will describe the text, style, and layout
• Success criteria:
• For at least one test phrase like “Real courage is doing the right thing when no one applauds you”,
7 out of 10 generations should be spelled correctly and totally readable. We would expect better for short phrases.
Single-Image UI Parameters
In ComfyUI, I need these parameters exposed in an easy-to-adjust way:
• Phrase text (FAITH OVER FEAR).
• Seed (integer).
• Steps (if applicable).
• CFG / guidance scale.
• Model selection (if you include more than one Flux checkpoint or LoRA).
• Number of variations (Default 1 options 1-6)
Don’t worry about fancy UI — just make sure a normal ComfyUI user (me) can see and tweak these easily in the node parameters.
Deliverables
ComfyUI Workflows (JSON)
At minimum:
1. Single-image workflow
o JSON file (e.g., ArtGeyser_Flux_Text_v1.json)
o Uses one phrase input and allows for 1 or multiple variations of the PNG.
Documentation
A short README.txt in plain English, containing:
1. Environment setup you actually used
o Python version
o Torch version (include CUDA tag, e.g. 2.3.0+cu121)
o Exact location of ComfyUI folder and venv
o List of installed node packs and versions (Impact Pack, etc.)
2. Model info
o Which Flux Schnell checkpoint(s) you used
o Download links (or at least release names / hashes)
o Where to place them on my system (e.g., models/checkpoints/FluxSchnell-Whatever.safetensors)
3. How to run the workflow
o Which JSON to load
o Which nodes to edit to change:
phrase text
background mode
embellishments
o Where the output PNGs are saved.
Acceptance Criteria (what I will use to sign off)
You can test with the phrase:
Real courage is doing the right thing when no one applauds you. Charlie Kirk
No money’s will be released until the job is accepted. I’ll accept the job when:
1. ComfyUI starts cleanly with no CUDA or node errors.
2. The workflow loads without missing nodes.
3. Running the workflow produces:
o A 1500×1500 (or 1024×1024) PNG.at least 7/10 images with correct spelling.
4. Embellishments appear in sensible positions.
5. I can change the phrase text to something like:
o GRACE UPON GRACE
o BE STILL AND KNOW
and re-run, without touching any wiring.
6. Nice-to-Have (Not required for this milestone, but helpful)
• A simple batch script or documented method to:
o Pass a list of phrases (from a text file or CSV)
o Run the ComfyUI workflow for each entry
o Save outputs with phrase-based filenames.
• Suggestions for:
o Better text LoRAs that you have actually tested with Flux Schnell or the model used.
o Best settings for stable text (CFG, steps, negative prompts).
________________________________________
7. What I Care About Most
1. Stability – no “it works only on my machine” setups. It MUST function correctly on my computer.
2. Text correctness
3. Clean, legible, sellable style – not grungy horror AI artifacts.
4. Clear documentation – so that my AI assistant (ChatGPT) and I can expand, batch, and automate this later.
If there are any questions please reach out to me. If you want to place a bid put the phrase “Peanut Butter” somewhere in the first sentence. That shows me that you actually have read this and are serious.
Thank you very much.
Frank Sousa
I run a design brand that sells high-quality word-art (script + decorative elements) on Etsy and other platforms. Samples can be viewed at this URL: https://inyobus.com/samples
I need a qualified ComfyUI developer to build a workflow that reliably generates sellable word-art images from text phrases, probably using Flux Schnell (or another model that would work well) with good spelling and consistent style.
Environment & Constraints
2. OS / Hardware
• Windows 11
• GPU: NVIDIA RTX 4060 Ti 16gb
• Plenty of disk space
3. Existing ComfyUI install
• Main install folder:
C:\Art Geyser\0 ComfyUI Engine
• Uses a virtualenv at C:\Art Geyser\0 ComfyUI Engine\venv
• ComfyUI-Manager is already present and used.
4. Target Models & Node Packs
Base model. Through my research it seems that Flux Schnell is a good suggested model for text, unless you have another suggestion that:
o works on RTX 4060 Ti 16gb
o has good text rendering capabilities. (Most important)
o is appropriate for typographic / sign / logo style images.
If Flux Schnell truly cannot be made to work reliably on this GPU, propose a close alternative with similar performance and typography quality, and document the choice.
Node packs / custom nodes
Please install & configure only what is needed, but at minimum:
1. Any Flux-specific loader / node pack required to run Flux Schnell, or whatever alternate model is used.
If additional nodes are needed (ControlNet, IPAdapter, etc.), list them explicitly and verify they are compatible with the versions you choose.
5. Functional Requirements – What the Workflow Must Do
High-level goal
Given a phrase like:
“FAITH OVER FEAR”
or a phrase like “Real courage is doing the right thing when no one applauds you” with a subline like an author’s quote “Charlie Kirk” the workflow should generate a word-art image like the samples I provide.
• Correct spelling at least 75–90% of the time.
• A consistent “sellable” style — clean, bold, legible typography.
• Embellishments like flourishes, borders, watercolor backgrounds hearts, dots or anything other relevant artistic embellishment.
This is not about random art; it’s about building a reliable typography engine that creates word art. These images can be used in a variety of products such as mugs, t-shirts, pillows and so on.
Image specs
Default output (can be configurable in the workflow):
• Size: 1500 × 1500 px (or 1024×1024 if Flux strongly prefers it, but document that)
• Mode: RGB
• Background:
o Mode A (default for now): pure white (#FFFFFF)
o Mode B (added from the prompt): light neutral / farmhouse-style textured background (beige, wood, etc, floral style, or any other background style described in the prompt. (most of the time not used.)
• File format: PNG saved into a configurable folder, with filename including the first three words in the phrase slug and seed (e.g., faith_over_fear.png). When there will be multiple phrases that are the same we will let windows number them (1)(2) etc.
6. Model-Driven Text
Flux Schnell (or similar) directly creates the letterforms and style.
• You may use a text-accuracy LoRA (like Texta or equivalent) if compatible with Flux or your chosen model.
• The prompt will describe the text, style, and layout
• Success criteria:
• For at least one test phrase like “Real courage is doing the right thing when no one applauds you”,
7 out of 10 generations should be spelled correctly and totally readable. We would expect better for short phrases.
Single-Image UI Parameters
In ComfyUI, I need these parameters exposed in an easy-to-adjust way:
• Phrase text (FAITH OVER FEAR).
• Seed (integer).
• Steps (if applicable).
• CFG / guidance scale.
• Model selection (if you include more than one Flux checkpoint or LoRA).
• Number of variations (Default 1 options 1-6)
Don’t worry about fancy UI — just make sure a normal ComfyUI user (me) can see and tweak these easily in the node parameters.
Deliverables
ComfyUI Workflows (JSON)
At minimum:
1. Single-image workflow
o JSON file (e.g., ArtGeyser_Flux_Text_v1.json)
o Uses one phrase input and allows for 1 or multiple variations of the PNG.
Documentation
A short README.txt in plain English, containing:
1. Environment setup you actually used
o Python version
o Torch version (include CUDA tag, e.g. 2.3.0+cu121)
o Exact location of ComfyUI folder and venv
o List of installed node packs and versions (Impact Pack, etc.)
2. Model info
o Which Flux Schnell checkpoint(s) you used
o Download links (or at least release names / hashes)
o Where to place them on my system (e.g., models/checkpoints/FluxSchnell-Whatever.safetensors)
3. How to run the workflow
o Which JSON to load
o Which nodes to edit to change:
phrase text
background mode
embellishments
o Where the output PNGs are saved.
Acceptance Criteria (what I will use to sign off)
You can test with the phrase:
Real courage is doing the right thing when no one applauds you. Charlie Kirk
No money’s will be released until the job is accepted. I’ll accept the job when:
1. ComfyUI starts cleanly with no CUDA or node errors.
2. The workflow loads without missing nodes.
3. Running the workflow produces:
o A 1500×1500 (or 1024×1024) PNG.at least 7/10 images with correct spelling.
4. Embellishments appear in sensible positions.
5. I can change the phrase text to something like:
o GRACE UPON GRACE
o BE STILL AND KNOW
and re-run, without touching any wiring.
6. Nice-to-Have (Not required for this milestone, but helpful)
• A simple batch script or documented method to:
o Pass a list of phrases (from a text file or CSV)
o Run the ComfyUI workflow for each entry
o Save outputs with phrase-based filenames.
• Suggestions for:
o Better text LoRAs that you have actually tested with Flux Schnell or the model used.
o Best settings for stable text (CFG, steps, negative prompts).
________________________________________
7. What I Care About Most
1. Stability – no “it works only on my machine” setups. It MUST function correctly on my computer.
2. Text correctness
3. Clean, legible, sellable style – not grungy horror AI artifacts.
4. Clear documentation – so that my AI assistant (ChatGPT) and I can expand, batch, and automate this later.
If there are any questions please reach out to me. If you want to place a bid put the phrase “Peanut Butter” somewhere in the first sentence. That shows me that you actually have read this and are serious.
Thank you very much.
Frank Sousa