Most FLUX models give you raw generation power. Fluxgram V1.0 gives you something more specific: photorealistic human characters with consistent skin rendering. It's a FLUX LoRA fine-tuned specifically for realism, skin fix, and character coherence — available via ModelsLab's API with a single model ID swap.
If you've been using base FLUX models and struggling with skin texture artifacts, inconsistent character faces, or that uncanny-valley realism problem, Fluxgram is worth testing. This guide covers the API integration, parameters, and real code examples.
What Is Fluxgram V1.0?
Fluxgram V1.0 (fluxgram-v1-0) is a FLUX LoRA model published by ModelsLab. LoRA stands for Low-Rank Adaptation — it's a fine-tuning technique that adjusts specific weights in the base FLUX model without retraining from scratch. The result is a model that preserves FLUX's strengths (prompt understanding, composition, resolution) while specializing in realistic character generation.
Key characteristics:
- Realism focus: Trained specifically on photorealistic human subjects
- Skin fix: Addresses common FLUX artifacts in facial and body skin rendering
- Character consistency: Better coherence for repeated character generation
- FLUX base: Inherits FLUX's architecture — high resolution, strong composition, detailed prompts
The model is available through ModelsLab's v6/images/text2img community endpoint. Cost: $0.0047 per image (free on premium plans).
API Setup
Get your API key from the ModelsLab dashboard. Then you're one API call away from generating with Fluxgram.
Basic Text-to-Image Request
import requests
import json
API_KEY = "your_modelslab_api_key"
payload = {
"key": API_KEY,
"model_id": "fluxgram-v1-0",
"prompt": "portrait of a 30-year-old woman, natural lighting, sharp facial features, professional photography, f/2.8, bokeh background",
"negative_prompt": "cartoon, illustration, painting, low quality, blurry, distorted, deformed skin",
"width": "1024",
"height": "1024",
"samples": 1,
"num_inference_steps": 30,
"safety_checker": "no",
"enhance_prompt": "yes",
"seed": None,
"webhook": None,
"track_id": None
}
response = requests.post(
"https://modelslab.com/api/v6/images/text2img",
headers={"Content-Type": "application/json"},
data=json.dumps(payload)
)
result = response.json()
print(result)
A successful synchronous response returns image URLs in result['output']. If the response status is processing, use the fetch_result endpoint with the returned ID.
Handling Async Responses
Fluxgram generations at high resolution may return a processing status with an ID. Here's how to poll for the result:
import requests
import time
API_KEY = "your_modelslab_api_key"
def generate_and_fetch(prompt: str) -> list[str]:
payload = {
"key": API_KEY,
"model_id": "fluxgram-v1-0",
"prompt": prompt,
"negative_prompt": "cartoon, low quality, deformed skin",
"width": "1024",
"height": "1024",
"samples": 1,
"num_inference_steps": 30,
}
resp = requests.post(
"https://modelslab.com/api/v6/images/text2img",
headers={"Content-Type": "application/json"},
json=payload
)
data = resp.json()
if data["status"] == "success":
return data["output"]
if data["status"] == "processing":
fetch_id = data["id"]
# Poll for result
for _ in range(30):
time.sleep(5)
fetch_resp = requests.post(
"https://modelslab.com/api/v6/images/fetch",
json={"key": API_KEY, "request_id": fetch_id}
)
fetch_data = fetch_resp.json()
if fetch_data["status"] == "success":
return fetch_data["output"]
raise TimeoutError("Generation timed out after 150s")
raise Exception(f"Unexpected status: {data['status']}")
# Usage
images = generate_and_fetch(
"headshot portrait, professional businessman, studio lighting, photorealistic, 8k"
)
print(images)
Key Parameters for Fluxgram
Not all parameters are equal for realism-focused generation. Here's what matters most for Fluxgram:
Inference Steps (num_inference_steps)
Range: 20–50. For Fluxgram's realism LoRA, 28–35 steps hits the sweet spot. Below 20 and you lose skin detail; above 40 gives diminishing returns with higher latency.
Resolution
Fluxgram performs best at 1024×1024 or portrait aspect ratios like 768×1152. The skin-fix LoRA was trained on high-resolution human subjects — lower resolutions don't fully activate the fine-tuning.
Negative Prompt
For realistic character generation, these negatives consistently improve output quality:
"cartoon, illustration, anime, painting, 3d render, cgi, deformed skin, plastic skin,
low quality, blurry, watermark, text, duplicate, morbid, disfigured"Enhance Prompt (enhance_prompt)
Set to "yes" for character prompts. ModelsLab's prompt enhancement adds photography terms and quality tokens that Fluxgram's realism training responds well to.
Prompt Strategies for Best Results
Fluxgram responds strongly to photography-style prompts. Think in terms of camera specs, lighting setups, and subject descriptions rather than artistic style keywords:
# Strong prompts for Fluxgram
prompts = [
# Professional portraits
"professional headshot, female executive, 40s, sharp features, studio lighting, Canon EOS R5, f/1.8 prime lens, neutral background",
# Lifestyle/editorial
"candid street photography, young man reading in cafe, natural light from window, warm tones, shallow depth of field",
# Product modeling
"model wearing casual denim jacket, outdoor urban setting, golden hour lighting, photorealistic, high detail",
# Environmental portraits
"scientist in lab coat examining sample, fluorescent lighting, detailed facial expression, documentary style",
]
Avoid vague quality tokens like "masterpiece" or "ultra high quality" — Fluxgram's LoRA does the quality work. Specificity in lighting and context beats generic quality boosters.
Use Cases Where Fluxgram Shines
1. Product Photography with Human Models
E-commerce teams use Fluxgram for mock product shots with human models — consistent skin rendering means the same prompt reliably produces brand-consistent visuals at scale.
2. Avatar and Profile Generation
Apps generating professional profile photos, avatar systems, or virtual personas benefit from Fluxgram's character consistency. The realism LoRA reduces the artifact rate that makes generated portraits look clearly AI-generated.
3. Content Creation Pipelines
Editorial content, blog post headers, social media assets — any pipeline that needs human subjects at volume. At $0.0047/image (or free on premium), running 1,000 images costs under $5.
4. Fine-Tuning Base Dataset
Fluxgram-generated images make reasonable training data for downstream LoRAs. The consistent skin rendering and realistic proportions are better base material than raw FLUX output for character-specific fine-tunes.
Comparing Base FLUX vs Fluxgram
The difference shows most in edge cases: unusual lighting angles, aged faces, hands near faces, and complex textures like stubble or fine hair. Base FLUX handles these cases reasonably; Fluxgram handles them more consistently because the LoRA was trained specifically to fix these weak spots.
For non-character content — landscapes, objects, abstract imagery — base FLUX models are the better choice. Fluxgram's fine-tuning is character-specific and doesn't add value for non-human subjects.
Pricing
Fluxgram uses the standard ModelsLab community model pricing:
- Pay-as-you-go: $0.0047 per image
- Premium plans: Image generation included (effectively $0 per image)
- No separate LoRA licensing fee — the model access is bundled with your API key
For high-volume applications (>10,000 images/month), the premium plan math is straightforward: $47+ in PAYG costs vs. a fixed monthly plan.
Getting Started
The quickest path to testing Fluxgram:
- Create a free ModelsLab account at modelslab.com
- Get your API key from the dashboard settings
- Run the basic text-to-image example above with
"model_id": "fluxgram-v1-0" - Compare against a base FLUX model (
"model_id": "flux") on the same prompt
The ModelsLab Fluxgram playground also lets you test prompts without writing code — useful for prompt iteration before wiring up your pipeline.
Browse all available FLUX and custom LoRA models in the ModelsLab model library.