A brokerage with 80 agents has 80 different headshot styles. One agent used a smartphone in natural light. Another paid a photographer five years ago. A third joined last week and submitted a photo taken at a family event. The company website looks like a random collection of portraits rather than a team. The brand signal is incoherent.
Fixing this manually means booking a photographer, coordinating 80 schedules, paying per session, and repeating every time a new agent joins or a photo needs refreshing. At $150-300 per professional headshot, a 80-agent brokerage is looking at $12,000-$24,000 just to get everyone consistent. Then doing it again in two years.
A realtor headshot API solves this with one POST request per agent. Any input photo goes in: selfie, event photo, off-brand studio portrait. A brokerage-grade headshot comes out: standardized background, lighting, framing, and attire. The result is consistent across every agent regardless of what they submitted.
TL;DR: A FaceDetect + BgRemove + HeadshotGen pipeline produces a standardized portrait in 2.1 seconds at $0.012 per image. Runflow includes Sentinel, a built-in quality assurance layer that automatically filters bad outputs before delivery. No manual QA. No rejected batches reaching the client.
| Stack | Infra /mo | AI team | Total cost | Revenue | Margin |
|---|---|---|---|---|---|
Runflow 10% volume discount applied | $2.7K | $0 | $2.7K | $20K | 87% |
Cloud API + manual QA similar pricing · no auto-QA · part-time engineer needed | $3.0K | ~$5K | $8.0K | $20K | 60% |
Self-hosted GPU raw compute · full-time AI engineer required | $400 | $12K | $12K | $20K | 38% |
Runflow Sentinel — built-in quality control layer that automatically detects and discards failed or low-quality outputs before delivery. You only pay for images that pass QA. No engineer needed to babysit the pipeline.
Pricing based on Runflow published rates (June 2026) with automatic volume discounts. Revenue column is illustrative — actual client pricing varies by vertical and contract size. GPU self-hosted estimate uses $0.04/img raw compute cost.
The market: who needs this and why now
There are approximately 1.5 million active real estate agents in the United States. Most work under a brokerage brand. Every brokerage has an agent roster page. Almost none of them have a consistent visual identity across their agent photos.
The strongest buyers are brokerage CRM platforms: kvCORE, BoomTown, Follow Up Boss, Sierra Interactive. These platforms manage agent profiles, listings, and client communications for brokerages with 50 to 5,000 agents. If they add a headshot standardization feature, every brokerage on their platform gets it automatically. One integration touches thousands of brokerages simultaneously.
Franchise brands with strict visual identity requirements are the second tier: Keller Williams, RE/MAX, Coldwell Banker. They mandate consistent agent photos in their brand guidelines but have no scalable way to enforce them. A headshot API built into their technology stack enforces the standard automatically.
How the pipeline works
The pipeline runs six nodes. LoadImage accepts any portrait format: JPEG, PNG, WebP, minimum 300x300 pixels. FaceDetect locates the face, aligns it, and normalizes the head position and tilt. BgRemove segments the subject from the background with pixel-level precision. HeadshotGen relights the face and composites it onto the brokerage background template with the specified lighting style. Sentinel runs quality assurance on the output. SaveImage delivers the final file.
The brokerage background template is configurable: background color, lighting temperature, shoulder crop ratio, and output dimensions. Set it once per brokerage. Every agent processed through that configuration gets the same visual treatment regardless of their input photo.
Sentinel: quality control built into the pipeline
At brokerage scale, bad outputs are inevitable without a quality gate. Low-resolution inputs produce artifacts. Extreme angles fool the face detector. Heavy background patterns bleed through the mask. Without a filter, some percentage of outputs reach the client with visible defects.
Sentinel is Runflow's built-in quality assurance layer. It runs after HeadshotGen and before SaveImage. It scores each output against a set of quality criteria: face completeness, background uniformity, edge artifact detection, and overall composition score. Outputs that fall below the confidence threshold are flagged automatically. The platform receives a rejection signal with the specific failure reason rather than a defective image.
In practice this means a brokerage platform can process 500 agent photos in a batch overnight and receive 490 approved results plus 10 flagged cases that need a better input photo. No human reviewer scanned the batch. No defective headshots reached the agent profile page. The platform prompts the 10 flagged agents to resubmit with a better photo.
| Stack | Infra /mo | AI team | Total cost | Revenue | Margin |
|---|---|---|---|---|---|
Runflow 10% volume discount applied | $2.7K | $0 | $2.7K | $20K | 87% |
Cloud API + manual QA similar pricing · no auto-QA · part-time engineer needed | $3.0K | ~$5K | $8.0K | $20K | 60% |
Self-hosted GPU raw compute · full-time AI engineer required | $400 | $12K | $12K | $20K | 38% |
Runflow Sentinel — built-in quality control layer that automatically detects and discards failed or low-quality outputs before delivery. You only pay for images that pass QA. No engineer needed to babysit the pipeline.
Pricing based on Runflow published rates (June 2026) with automatic volume discounts. Revenue column is illustrative — actual client pricing varies by vertical and contract size. GPU self-hosted estimate uses $0.04/img raw compute cost.
Cost at every scale: API vs self-built
The build-vs-buy question depends on volume and team capacity. Below are the real numbers at three scales.
| Under 1,000 imgs/mo | 10,000 imgs/mo | 100,000 imgs/mo | |
|---|---|---|---|
| Runflow API cost | $12 | $120 | $1,200 |
| Self-built + full-time ML engineer ($10K/mo + $1,200 GPU) | $11,200 | $11,200 | $11,200 |
| Self-built + part-time contractor ($4K/mo + $600 GPU) | $4,600 | $4,600 | $4,600 |
| Time to first output | 1-2 days | 1-2 days | 1-2 days |
| Time to first output (self-built) | 3-5 months | 3-5 months | 3-5 months |
| Sentinel QA included | Yes | Yes | Yes |
The crossover point where self-building with a full-time engineer becomes cheaper than the API is around 1 million images per month. Below that threshold, the API wins on every dimension: cost, speed, and quality. Most brokerage CRM platforms processing headshots for their customer base will never reach that volume.
The part-time contractor path looks cheaper at 100K images, but it comes with hidden costs. A contractor working 20 hours per week cannot maintain a production pipeline, respond to incidents, or iterate on model quality. The moment something breaks at 2am, the platform has no coverage. The API includes uptime SLA, support, and model updates. The contractor does not.
What you can charge and the margin you keep
The API cost is $0.012 per image. The question is what the market will pay for the output.
| Selling model | Price per headshot | API cost | Gross margin |
|---|---|---|---|
| Premium brokerage service (white-label) | $15.00 | $0.012 | 99.9% |
| CRM platform add-on (bulk brokerage) | $5.00 | $0.012 | 99.8% |
| SaaS subscription ($49/mo per agent, 2 headshots/yr) | $2.04 per use | $0.012 | 99.4% |
| Consumer AI headshot app (self-service) | $0.99 | $0.012 | 98.8% |
The gross margin on the API cost alone is above 98% across every pricing model. The practical limit on profitability is not the API cost but the overhead of running the platform: hosting, support, sales, and customer success. At any scale above a few thousand images per month, the unit economics are exceptionally strong.
The premium brokerage service model is the highest-value positioning. A white-label headshot service that charges $15 per portrait and delivers it in two seconds is competing against photographers charging $150-300 with a two-week lead time. The customer does not need convincing. They need to see the output quality once.
The CRM platform add-on model is the highest-volume path. A platform with 10,000 active brokerages adding a headshot feature at $5 per agent photo can generate $50,000+ in monthly recurring revenue from a feature that costs $120 per month in API calls at that volume. The ratio is not a typo.
Implementation: the integration is one call
The platform sends a base64-encoded photo and the brokerage configuration ID. The API returns a standardized headshot with a Sentinel quality score. No model management. No GPU provisioning. No background removal service to stitch together separately.
import requests, base64
with open("agent_photo.jpg", "rb") as f:
image_b64 = base64.b64encode(f.read()).decode()
response = requests.post(
"https://api.runflow.io/v1/workflows/realtor-headshot",
headers={"Authorization": "Bearer YOUR_API_KEY"},
json={
"image": image_b64,
"brokerage_config": "coldwell-banker-west"
}
)
result = response.json()
print(result["headshot_url"]) # standardized output URL
print(result["sentinel_score"]) # 0.0-1.0 quality score
print(result["sentinel_passed"]) # True / FalseThe sentinel_passed field is the QA gate. If False, the platform prompts the agent to resubmit. The sentinel_score gives the platform flexibility to set its own threshold: a consumer app might accept 0.80+, while a premium white-label service requiring publication-quality results might require 0.92+.
Who is not a good fit for this
HeadshotPro and similar direct-to-consumer AI headshot services have already built their own pipelines and own the consumer market. This API is not for building a competing consumer product. The buyers are platforms that want headshot generation as a feature, not as their core product.
Brokerages that want fully custom photorealistic outputs with specific wardrobe choices, studio backgrounds, or high-end retouching are better served by professional photography. The API excels at consistent standardization at volume, not at bespoke artistic direction.
The sweet spot is platforms processing 50 to 50,000 agent headshots per month that need consistency, speed, and no manual overhead. At those volumes, the economics are clear and the operational complexity is negligible.
What to build beyond the headshot
The same pipeline that standardizes agent headshots can process other real estate photography use cases with minor configuration changes. Listing photo enhancement, virtual staging, and day-to-dusk conversion all run on the same infrastructure. A CRM platform that starts with headshot standardization has a natural expansion path into full listing photo automation without rebuilding the integration.
The agent profile is the foundation. Once the platform owns the agent photo workflow, it has a logical reason to own the listing photo workflow. The headshot API is the entry point. The real estate photography stack is the expansion. Both run through the same API key.
The integration path is straightforward for any CRM platform that already handles agent profile data. The headshot API sits between the photo upload step and the profile publish step. The agent submits any photo. The platform calls the API. Sentinel validates the output. If the quality score passes, the standardized headshot publishes automatically. If not, the platform sends the agent a prompt to resubmit with a better photo. The entire workflow is automated. No coordinator manages the photo collection process. No reviewer checks outputs before they go live. The brokerage brand stays consistent without any ongoing manual effort from the platform team.