Manually writing alt-texts for thousands of images is an unrealistic task. A media library of 50,000 images demands hundreds of human-hours, and the quality often suffers due to editor fatigue. Our AI system processes such volumes in one hour with 94% accuracy—10 times faster and 20% cheaper than manual labor. We automate this process using vision-language models, delivering quality close to editorial standards at a fraction of the time.
According to internal tests, the system achieves 94% alignment with editorial standards on a sample of 1000 images.
Why automating alt-texts is a necessity
Without alt-texts, your content remains invisible to screen readers and ranks poorly in search results. Manually handling even 10,000 images requires weeks of a copywriter's work, and consistency suffers. Our system solves both problems: it generates accurate, SEO-optimized descriptions automatically, taking into account the page context and brand guidelines.
How we achieve 94% accuracy
Accuracy is the result of contextual prompting: the system receives the page title, category, and surrounding text. For each client, we configure description rules—for example, mandatory brand mentions or exclusion of certain objects. Regular A/B tests compare machine-generated descriptions against editorial ones, and we adjust the model when deviations occur.
Want to see how it works on your data? Get a test run for 100 images—just contact us.
How we build the system: stack and approach
For alt-text generation, we use a combination of vision-language models:
-
GPT-4V / GPT-4o — maximum accuracy, context understanding, support for complex scenes.
-
LLaVA 1.6 / InternVL2 — self-hosted option for strict confidentiality requirements.
-
BLIP-2 — lightweight model for high-frequency generation (batch up to 500 images/min).
Integration is done via REST API with CMS (WordPress, Contentful, Strapi) or S3/GCS buckets. Operation modes: scheduled bulk processing or real-time hook on image upload. Prompts are customized to the brand style—what to include (objects, colors, actions) and what to ignore. To optimize quality, we apply fine-tuning with LoRA and control p99 latency at <150ms.
Comparison: automated generation vs. manual work
| Criteria |
Manual generation (copywriter) |
Our AI system |
| Speed |
1–2 minutes per image |
100–500 images/min (batch) |
| Description accuracy |
~95% (average) |
~94% (vs human benchmark) |
| Consistency |
Depends on the writer |
Uniform style across all descriptions |
| Scalability |
Linear cost growth |
Nearly constant cost at large volumes |
| Language support |
Depends on linguist |
50+ languages out of the box |
Model comparison: when to use what
| Model |
When to choose |
| GPT-4V/GPT-4o |
Maximum accuracy, complex scenes, no data transfer restrictions |
| LLaVA 1.6 / InternVL2 |
Self-hosted, confidentiality, control over infrastructure |
| BLIP-2 |
High throughput, bulk processing, low cost per million tokens |
How to set up prompts for your brand?
The system supports templates with variables: brand name, color palette, mandatory objects. For an online clothing store, you could specify: "Mention the brand at the beginning, describe the color, style, and material. Avoid personal opinions." Ready prompts are tested on a sample of 200 images before launch.
Work process: from audit to deployment
- Analytics: we study your media library, page structure, and description requirements.
- Design: select the model, design the pipeline, configure prompts.
- Implementation: integrate with CMS via API, configure batch and real-time modes.
- Testing: compare against reference descriptions, adjust prompts until target quality is achieved.
- Deployment: roll out to production, monitor quality via A/B tests.
What's included
- Documentation: solution architecture, operation instructions, API description.
- Access: to models (cloud or self-hosted), to the pipeline, to the monitoring dashboard.
- Training: 1–2 hour workshop for content managers and developers.
- Support: 2 weeks post-launch support for tuning and optimization.
Timeline and cost
A typical project takes 1 to 2 weeks from the moment we receive access. The cost is calculated individually, based on media library size and integration complexity. We don't hide pricing: contact us to get a preliminary estimate within one day.
Our team has completed 12 content automation projects for major retailers—that's 5 years of AI experience and hundreds of thousands of images processed. We guarantee data confidentiality when using self-hosted models and compliance with WCAG 2.1 AA. All system components are tested and validated in production.
Contact us to discuss your project. Get a consultation on implementation today.
Generative AI Development: From Prompt to Production API
We often receive a task "generate a product image" — on the surface it seems simple. But behind this lies a choice between dozens of models, configuring the inference pipeline, manually solving consistency issues, integrating into the product backend, and answering why the model generates hands with six fingers in staging but not in production. Let's break down the directions we work with.
Image Generation: From Prompt to Production API
The current landscape includes FLUX.1 [dev/schnell/pro] from Black Forest Labs and Stable Diffusion 3.5. FLUX.1 [schnell] takes 4 steps instead of 20–50 for SDXL — 5–12 times faster — while maintaining higher quality. On an A100 80GB — 1.2–1.8 s per 1024×1024 image at batch_size=4.
A typical deployment issue: FLUX.1 [dev] requires 24+ GB VRAM in fp16. On A10G 24GB it fits tightly; at batch_size>1 — OOM. Solution: torch_dtype=torch.bfloat16 + enable_model_cpu_offload() from diffusers, or quantization via bitsandbytes to NF4 — minimal quality drop, memory consumption drops to 12–14 GB.
ControlNet and IP-Adapter are key tools for production tasks where controllability is needed. ControlNet with Canny/Depth/Pose maps provides structural control. IP-Adapter (especially IP-Adapter-FaceID) allows transferring character identity to generations — this is the foundation for personalized content. More about ControlNet can be found on Wikipedia.
Case study: e-commerce photography. A retailer with 8000 SKUs needed lifestyle photos for each product. Pipeline: product segmentation (Segment Anything Model 2) → background removal → inpainting with FLUX.1 [dev] using product image as IP-Adapter reference → upscale via RealESRGAN_x4plus. The generation cost is negligible compared to professional photography, providing huge savings. Throughput — 200 images/hour on 2× A100. Our extensive experience from 30+ projects ensures we select the optimal model for your task — an evaluation can be obtained upfront.
Why Is Model Selection Only Half the Battle?
Fine-tuning for a Specific Style or Character
Dreambooth and LoRA are the standard for adapting to a specific visual style or object. LoRA trains in 2–4 hours on 20–30 reference images on a single A100. Rank 16–32 is usually sufficient for style; rank 64+ is needed for precise face reproduction.
A common mistake: training LoRA too long — the model overfits to references, losing the ability to vary. Sign: at cfg_scale=7, all images look like copy-paste of references. Solved by early stopping (usually 1500–2000 steps for 20 images) and prior_preservation_loss.
For deeper customization — full fine-tuning via diffusers + accelerate with FSDP on multiple GPUs. But that already takes 40–80 hours of training and requires a truly large dataset (1000+ images).
Comparison of Image Generation Approaches
| Model |
Speed (1024×1024, A100) |
Quality (CLIP score) |
Controllability (ControlNet, IP-Adapter) |
VRAM (fp16) |
| Stable Diffusion 3.5 |
2.0–3.5 s |
0.28–0.31 |
via ControlNet (allowed) |
16–20 GB |
| FLUX.1 [schnell] |
0.8–1.2 s |
0.30–0.33 |
limited (no ControlNet) |
12–14 GB (4‑step) |
| FLUX.1 [dev] |
3–5 s (50 steps) |
0.32–0.34 |
via IP-Adapter, ControlNet (adapter) |
24+ GB |
| Midjourney (API) |
5–10 s (queue) |
0.31–0.33 |
prompt + style reference |
not required |
Video Generation: Which Models Are Best?
| Model |
Availability |
Duration |
Resolution |
Controllability |
| Sora (OpenAI) |
API (limited) |
up to 60 s |
1080p |
prompt, image-to-video |
| Wan2.1 (Alibaba) |
open weights |
up to 81 frames |
720p |
prompt, I2V, V2V |
| CogVideoX-5B |
open weights |
6 s |
720p |
prompt, I2V |
| Kling 1.6 |
API |
up to 30 s |
1080p |
prompt, I2V |
| Mochi-1 |
open weights |
5.4 s |
480p |
prompt |
Open-weight video models still lag behind commercial ones in stability and length. Wan2.1 is the best choice for self-hosting: 14B parameters, runs on 2× A100, delivers acceptable quality for short clips.
The main pain of video generation is temporal consistency: the character changes clothing color at the third second, objects "drift." Partial solution — generation with motion_bucket_id and noise_aug_strength in Stable Video Diffusion, or using I2V (image-to-video) instead of pure text-to-video. As noted in VideoPoet research, consistency is achieved by training on long sequences.
AnimateDiff remains a working tool for short loops and motion effects on top of SD/FLUX. Not Sora, but deployable locally and predictable.
Music and Audio Generation
AudioCraft from Meta (MusicGen + AudioGen) is a production-ready stack for music generation. musicgen-large (3.3B) generates 30 s of music in ~8 s on A100. Control via text prompt and melody conditioning — you can specify a melody by humming.
Stable Audio Open from Stability AI is an alternative with length up to 47 s, better structural control (intro/verse/chorus). Deployment is similar: diffusers + FastAPI.
For voice-over and dubbing — ElevenLabs API or self-hosted XTTS v2 (see Speech AI service). For sound design and foley — AudioGen.
3D Generation: Current Practical State
3D generation has not yet reached the same maturity as 2D. But for specific tasks, tools are already working:
TripoSG and Shap-E — text/image-to-3D. Shap-E from OpenAI generates simple 3D meshes in seconds, but geometry is rough. TripoSG gives more detailed results but requires post-processing (remeshing, UV unwrapping).
Wonder3D and Zero123++ — 3D reconstruction from a single image. They work by generating multi-views (6–8 views) and then 3D reconstruction via NeuS or instant-ngp.
Gaussian Splatting (3DGS) — not generation, but reconstruction from a series of photos/videos. For product cards and real estate it's already production: 50–200 photos → 3DGS model in 15–30 min on RTX 4090 → interactive 3D viewer in browser.
What Infrastructure Is Needed for Generative AI Deployment?
Critical for generative models:
- Task queue — Celery + Redis or Ray Serve. Synchronous HTTP for image generation is unacceptable with >5 concurrent requests.
- Caching — similar prompts yield similar results. Semantic cache via embeddings (faiss + sentence-transformers) can reduce GPU load by 20–40%.
- Quality monitoring — CLIP score for text-image alignment, FID for evaluating generation distribution. Integrate into MLflow or Weights & Biases.
- Storage — generated images immediately to S3/MinIO, not on the inference server disk.
What's Included in the Deliverables
We take the project turnkey — from model selection to deployment and monitoring. The result includes:
- Model (or API integration) with performance benchmarks (latency p99, throughput).
- Pipeline documentation (prompt engineering guide, model card, dependency versions).
- Integration with your backend (REST/gRPC, queues).
- Configured monitoring (dashboards, alerts for quality drift).
- Training workshop for the team (2–4 hours).
- Warranty support for 3 months after launch — as part of our quality certificate.
We have completed 30+ projects in generative AI — this gives us the right to guarantee results.
How Is the Generative AI Development Process Structured?
- Analysis (1–2 days): audit of current architecture, clarification of use case, selection of models and success metrics. We evaluate the project free of charge.
- Proof of Concept (1–3 weeks): quick prototype on your data — to see real quality, not blog demos.
- Design (1–2 weeks): pipeline architecture, infrastructure (GPU cluster/API), A/B testing plan.
- Implementation and fine-tuning (4–12 weeks): development, LoRA/full fine-tuning, integration with queue and cache.
- Testing (1–2 weeks): load tests, metric validation, edge-case verification (negative scenarios).
- Deployment and monitoring (1–2 weeks): production deployment, monitoring setup, documentation.
What We Verify at the Proof of Concept Stage
- Alignment of expectations and actual generation quality (CLIP score, user study).
- Inference speed at different batch sizes and GPU types.
- Likelihood of toxic/incorrect generations — checking safety filters.
- Scalability: will the model handle peak load.
Timeline Estimates
Integration of a ready API (DALL·E 3, Midjourney API, Stability API) — 1–2 weeks. Self-hosted pipeline with fine-tuning — 6–12 weeks. Full platform with UI, queues and monitoring — 3–6 months. The specific cost is calculated individually after analyzing your scenario.
Contact us — order a consultation, and we will select the optimal architecture for your project. Get a preliminary cost and timeline estimate for free.
